DETAILED ACTIONS
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 02/07/2025. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 9-36 are rejected under 35 U.S.C. 103 as being unpatentable over
Heifets et al. (US 2019/0164021 A1, hereinafter Heifets,) and in view of Ragoza et al. “Protein-Ligand Scoring with Convolutional Neural Networks”, arXiv:1612.02751v1 [stat.ML] 8 Dec 2016, (hereinafter, Ragoza, IDS reference).
Regarding Claim 1, Heifets teaches,
A computer system for characterizing an interaction between a test compound and a target polymer, (Heifets, Figure 1, [0003] “the prediction the affinity of test
objects to target objects”. [0007] “target object (e.g., a polymer)”) the computer system comprising:
one or more processors (Heifets, figure 2a, [0048] the computer system 100 comprises a general processor 74”); and
memory addressable by the one or more processors, the memory storing at least one program for execution by the one or more processors, the at least one program comprising instructions (Heifetz, figure 2a, [0048] “the computer system 100 comprises a general processor 74. general memory 90 / 92 addressable by the general processing unit, the general memory storing at least one program 56 for execution by the at least one general processor”); for:
(A) obtaining a plurality of atomic coordinates for the target polymer, wherein the plurality of atomic coordinates comprises atomic coordinates for; (Heifets, figure 2a, step 208, [0007] “The at least one program comprises instructions for obtaining a set of three-dimensional coordinates {xi, ... , xN} for all or a portion of a target object ( e.g., a polymer having an active site). Each respective x, in {xi, ... , xN} is a three dimensional coordinate for an atom in a plurality of atoms in the target object”). at least 400 atoms (Heifets teaches in [0071] “the systems and methods of the present disclosure have no limitation on the size of the test objects 72 or training objects 66. For instance, in some embodiments, such objects are large polymer, such as antibodies [0061]” In some embodiments the target object is a polymer and there are ten or more, twenty or more, thirty or more, fifty or more, one hundred or more, between one hundred and one thousand, or less than 500 residues in the polymer”. a polymer with 100 or more residues means the chain is composed of 100 or more repeating units connected together. Each polymer unit contains multiple atoms depends on the length and types of polymers. Therefore, for example a polymer is a protein polymer on average contains 12-15 atoms and with 100 residues total atoms will be 1200, which is more than 400 atoms. This is known in the art.)
(B) obtaining a training dataset comprising a respective electronic description of each training compound in a plurality of training compounds, wherein the plurality of training compounds comprises at least 100 compounds, (Heifets, FIG. 1, 92, Object training library, Training object 66-1, Figure 2A, In step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”. In step 218 “The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 or more poses”. “ plurality of poses” reads on the electronic description of each training compound. Also see [0073])
each respective electronic description comprising:
(i) a corresponding positive pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first positive interaction score (Heifets, Figure 2a, 212,” The target object is a polymer and the spatial coordinates are an ensemble of three-dimensional coordinates of the polymer”. Figure 2A, step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”.[ 0139] Obtaining a Plurality of Scores from the Scorer (276) and Using a Score from the Convolutional Neural Network to Characterize the Test Object (278)”), and
(C) training at least a first model, wherein the first model has a first plurality of parameters (Heitfets, Figure 2A,step 214-216, “ Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses. The target object 58 is a polymer with an active site, the test object is a chemical compound, and the docking comprises docking the test object into the active site of the polymer” , and wherein the first plurality of parameters comprises more than 400 parameters, the training using, for each corresponding training compound in the plurality of training compounds, at least (Heifets, [0010] (i) sampling the test object, in a respective pose in the plurality of different poses, and a portion of the target object that is contact with the test object in the respective pose ( e.g., the active site of the target object in instance where the target object is a polymer) on a three-dimensional basis (e.g., grid basis) thereby forming a corresponding three dimensional space filling uniform honeycomb comprising a corresponding plurality of three-dimensional space filling polyhedral cells. For each respective three-dimensional cell in the corresponding plurality of three-dimensional cells, a voxel in the respective voxel map is populated based upon one or more properties (e.g., chemical properties such as distance, angle, atom type, charge and polarization, and surrounding stabilizing or destabilizing environmental factors) of the respective three-dimensional cell”. Figure 2A, step 214, “ The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 more poses.”. Therefore, for each respective pose corresponding plurality of parameters, therefore for 100 or more poses with multiple parameters accounted for more than 400 parameters.)
(i) a corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first positive interaction score of the corresponding training compound with respect to the target polymer (Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”. Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”) and
wherein at least an output of the first model is used, at least in part, to provide the characterization of the interaction between the test compound and the target polymer. (Heifets, Figure 2E, steps 282,284,286,288, 290, [0009] “The at least one program further comprises instructions for obtaining a plurality of scores from the scorer, where each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors (or a corresponding voxel map) into the input layer. The plurality of scores is used to determine a classification of the test object. In some embodiments, a weighted average of the plurality of scores is used to determine a classification of the test object”)
Heifets teaches incorrect poses when meaning that such poses do not represent true interactions or binding energy between the test object 72 (or training object 66) and the target object 58. The wrong interactions are example of non-binding interactions. see Heifets, [0073]”. However, Heifets is silent on negative score determination.
Heifets is silent on (ii) a corresponding negative pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first negative interaction score;
(ii) a corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the first plurality of parameters,
However, Ragoza teaches ii) a corresponding negative pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first negative interaction score;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
(ii) a corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the first plurality of parameters, (Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 2, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer (Heifets, [0090] In some embodiments, a structural protein-ligand
interaction fingerprint (SPLIF) score is generated for each pose of a given test object (or training object) to a target object and this SPLIF score is used as additional input into the underlying neural network or is individually encoded in
the voxel map”) is obtained by:
obtaining a corresponding positive voxel map of the corresponding training compound with respect to the target polymer in the corresponding positive pose, unfolding the corresponding positive voxel map into a corresponding positive vector (Heifets, Figure 2D, step 262, “Unfold each voxel map in the plurality of voxel maps into a corresponding vector, thereby creating a plurality of vectors, where each vector 22 in the plurality of vectors is the same size”)., and inputting the corresponding positive vector to a convolutional neural network thereby obtaining the corresponding positive score for the corresponding positive pose, wherein the convolutional neural network comprises more than 500 parameters (Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”. Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”) and
Heifets is silent on the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer
is obtained by:
obtaining a corresponding negative voxel map of the corresponding training compound with respect to the target polymer in the corresponding negative pose,
unfolding the corresponding negative voxel map into a corresponding negative vector, and inputting the corresponding negative vector to the convolutional neural network thereby obtaining the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer.
However, Ragoza teaches the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer ;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
is obtained by: obtaining a corresponding negative voxel map of the corresponding training compound with respect to the target polymer in the corresponding negative pose, unfolding the corresponding negative voxel map into a corresponding negative vector, and inputting the corresponding negative vector to the convolutional neural network thereby obtaining the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer. (Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 3, combination of Heifets and Ragoza teaches the computer system of claim 2,
Heifets further teaches the corresponding positive vector is a first one-dimensional vector, and the corresponding negative vector is a second one-dimensional vector. (Heifets, Figure 2D, [0110] Referring to element 262, each voxel map 40 is optionally unfolded into a corresponding vector, thereby creating a plurality of vectors, where each vector in the plurality of vectors is the same size. In some embodiments, each vector in the plurality of vectors is a one-dimensional vector (264).”)
Regarding Claim 4, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the first model is a first fully connected neural network (Heifets, Figure 2E, [0131] In some embodiments, the scorer 30 comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully connected layers feeds into the evaluation layer (272)”).
Regarding Claim 9, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches, wherein the corresponding first positive interaction score and each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer. (Heifets, [0090] In some embodiments, a structural protein-ligand interaction fingerprint (SPLIF) score is generated for each pose of a given test object (or training object) to a target
object and this SPLIF score is used as additional input into the underlying neural network or is individually encoded in the voxel map.).
Heifets is silent on the corresponding first negative interaction score each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer
However, Ragoza teaches the corresponding first negative interaction score each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer ;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 10, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches Wherein each respective electronic description in the training dataset further comprises a corresponding positive activity score for the corresponding positive pose of the corresponding training compound (Heifets, [0090] In some embodiments, a structural protein-ligand interaction fingerprint (SPLIF) score is generated for each pose of a given test object (or training object) to a target object and this SPLIF score is used as additional input into the underlying neural network or is individually encoded in the voxel map”) and
the training at least the first model (C) further comprises jointly training a second model with the first model, wherein the second model has a second plurality of parameters, the training further using, for each corresponding training compound in the plurality of training compounds, at least:(iii) the corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer as input to the second model, against the corresponding positive activity score of the corresponding training compound(Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”. Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”), and
wherein the second model provides an activity of the interaction between the test compound and the target polymer that is used with the output of the first model, at least in part, to provide the characterization of the interaction between the test compound and the target polymer(Heifets, Figure 2E, steps 282,284,286,288, 290, [0009] “The at least one program further comprises instructions for obtaining a plurality of scores from the scorer, where each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors (or a corresponding voxel map) into the input layer. The plurality of scores is used to determine a classification of the test object. In some embodiments, a weighted average of the plurality of scores is used to determine a classification of the test object”)
Heifets is silent on a corresponding negative activity score for the corresponding negative pose of the corresponding training compound
(iv) the corresponding negative score for the corresponding negative pose of the
corresponding training compound with respect to the target polymer as input to the
second model, against the corresponding negative activity score of the corresponding
training compound, thereby adjusting the second plurality of parameters,
However, Ragoza teaches a corresponding negative activity score for the corresponding negative pose of the corresponding training compound Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
(iv) the corresponding negative score for the corresponding negative pose of the
corresponding training compound with respect to the target polymer as input to the
second model, against the corresponding negative activity score of the corresponding
training compound, thereby adjusting the second plurality of parameters, (Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 11, combination of Heifets and Ragoza teaches the computer system of claim 10,
Heifets further teaches wherein the second model is a second fully connected neural network (Heifets, Figure 2E, [0131] In some embodiments, the scorer 30 comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully connected layers feeds into the evaluation layer (272)”).
Regarding Claim 12, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the characterization of the interaction between the test compound and the target polymer is a binary activity score. (Heifets, Figure 2A, 232 The characteristic of the atom is encoded in the voxel as a binary categorical variable”)
Regarding Claim 13, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein each respective electronic description in the training dataset further comprises a corresponding positive activity score for the corresponding positive pose of the corresponding training compound (Heifets, FIG. 1, 92, Object training library, Training object 66-1, Figure 2A, In step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”. In step 218 “The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 or more poses”. “plurality of poses” reads on the electronic description of each training compound. Also see [0073]) and
the training at least the first model (C) further comprises jointly training a second model with the first model, wherein the second model has a second plurality of parameters, the training further using, for each corresponding training compound in the plurality of training compounds, (Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”., and
at least:(iii) the corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer and the corresponding first positive interaction score as joint input to the second model, against the corresponding positive activity score of the corresponding training compound, (Heifets, Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”) and
thereby adjusting the second plurality of parameters, wherein the second model is used with the output of the first model, at least in part, to provide the characterization of the interaction between the test compound and the target polymer (Heifets, [0151], Errors in activity class assignments made by the neural network, as verified against the binding data 68, are then back-propagated through the weights of the neural network in order to train the neural network 24. For instance, the filter weights of respective filters in the convolutional layers 28 of the network are adjusted in such back-propagation. In an exemplary embodiment, the neural network 24 is trained against the errors in the activity class assignn1ents made by the network 24, in view of the binding data 68, by stochastic gradient descent with the AdaDelta adaptive learning method”).
Heifets is silent on a corresponding negative activity score for the corresponding negative pose of the corresponding training compound,(iv) the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer and the corresponding first negative interaction score as joint input to the second model, against the corresponding negative activity score of the corresponding training compound,
However, Ragoza teaches a corresponding negative activity score for the corresponding negative pose of the corresponding training compound, ;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
(iv) the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer and the corresponding first negative interaction score as joint input to the second model, against the corresponding negative activity score of the corresponding training compound, (Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 14, combination of Heifets and Ragoza teaches the computer system of claim 13,
Heifets further teaches wherein the corresponding positive activity score is a first binary activity score and the corresponding negative activity score is a second binary activity score. (Heifets, Figure 2A, 238 “The plurality of channels includes a first channel that is a binary categorical variable for presence of an atom in the test object and second channel that is a binary categorical variable for presence of an atom in the target object”).
Regarding Claim 15, combination of Heifets and Ragoza teaches the computer system of claim 14,
Heifets further teaches wherein the corresponding first binary activity score is assigned a value of 1 based on a measured activity of the corresponding compound against the target polymer, and the corresponding second binary activity score is assigned a value of 0. (Heifets, [0086], When a given atom type is found in the three-dimensional grid element corresponding to a given voxel, the channel for
that atom type within the given voxel is assigned a first value of the binary categorical variable, such as "1", and when the atom type is not found in the three-dimensional grid element corresponding to the given voxel, the channel for that atom type is assigned a second value of the binary categorical variable, such as "0" within the given voxel”).
Regarding Claim 16, combination of Heifets and Ragoza teaches the computer system of claim 13,
Heifets further teaches wherein the training of the first model is a regression task in which the first plurality of parameters is adjusted by back-propagation through a first associated loss function and the training of the second model is a classification task in which the second plurality of parameters is adjusted by back-propagation through a second associated loss function (Heifets, [0008] a multiple additive regression tree, a clustering algorithm, principal component analysis, a nearest neighbor analysis (…) or ensembles thereof. [0150] Training the Predictive Model. In some embodiments, where a deep neural network is implemented (e.g.,the convolutional neural network 24), the convolutional assessment module 20 is configured to train the network 24 to receive the geometric data input and to output a prediction (probability) of whether or not a given test object binds to a target object. t [0158] In an embodiment, the neural network 24 may optionally, where training data is labeled (e.g., with the binding data 68), tune the weights within the network 24 to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations. Various methods may be used to minimize error function, such as gradient descent methods, which may include, but are not limited to, log-loss, sum of squares error, hinge-loss methods”. [0151] For instance, the filter weights of respective filters in the convolutional layers 28 of the network are adjusted in such back-propagation”),
Regarding Claim 17, combination of Heifets and Ragoza teaches the computer system of claim 16,
Heifets further teaches, wherein the corresponding first positive interaction score and, and the corresponding positive activity score is a first binary activity score and (Heifets, [0008] a multiple additive regression tree, a clustering algorithm, principal component analysis, a nearest neighbor analysis (…) or ensembles thereof. [0150] Training the Predictive Model. In some embodiments, where a deep neural network is implemented (e.g.,the convolutional neural network 24), the convolutional assessment module 20 is configured to train the network 24 to receive the geometric data input and to output a prediction (probability) of whether or not a given test object binds to a target object. t [0158] In an embodiment, the neural network 24 may optionally, where training data is labeled (e.g., with the binding data 68), tune the weights within the network 24 to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations. Various methods may be used to minimize error function, such as gradient descent methods, which may include, but are not limited to, log-loss, sum of squares error, hinge-loss methods”. [0151] For instance, the filter weights of respective filters in the convolutional layers 28 of the network are adjusted in such back-propagation”)
Heifets is silent on the corresponding first negative interaction score each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer and the corresponding negative activity score is a second binary activity score.
However, Ragoza teaches the corresponding first negative interaction score each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer ;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”) the corresponding negative activity score is a second binary activity score. (Ragoza, Page 9, op paragraph, This layer can process either standard molecular data files, which are read using OpenBabel,44 or a compact, custom binary gninatypes file that contains only the atomic coordinates and pre-processed atom type information”)
Regarding Claim 18, combination of Heifets and Ragoza teaches the computer system of claim 16,
Heifets further teaches, he computer system of claim 16, wherein the first associated loss function is a mean squared error loss function, a mean absolute error loss function, a Huber loss function, a Log-Cosh loss function, or a quantile loss function, and the second associated loss function is a binary cross entropy loss function, a hinge loss function, or a squared hinged loss function. (Heifets, [0158] In an embodiment, the neural network 24 may optionally, where training data is labeled (e.g., with the binding data 68), tune the weights within the network 24 to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations. Various methods may be used to minimize error function, such as gradient descent methods, which may include, but are not limited to, log-loss, sum of squares error, hinge-loss methods”. [0151] For instance, the filter weights of respective filters in the convolutional layers 28 of the network are adjusted in such back-propagation”),
Regarding Claim 19, combination of Heifets and Ragoza teaches the computer system of claim 13,
Heifets further teaches wherein the second model is a second fully connected neural network. (Heifets, Figure 2E, [0131] In some embodiments, the scorer 30 comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully connected layers feeds into the evaluation layer (272)”).
Regarding Claim 20, Heifets teaches,
The computer system of claim 1,
Wherein each respective electronic description in the training dataset further comprises a corresponding second positive interaction score for the corresponding positive pose of the corresponding training compound. Heifets, FIG. 1, 92, Object training library, Training object 66-1, Figure 2A, In step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”. In step 218 “The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 or more poses”. “ plurality of poses” reads on the electronic description of each training compound. Also see [0073])
, each respective electronic description in the training dataset further comprises a corresponding positive activity score for the corresponding positive pose of the corresponding training compound (Heifets, [0090] In some embodiments, a structural protein-ligand interaction fingerprint (SPLIF) score is generated for each pose of a given test object (or training object) to a target object and this SPLIF score is used as additional input into the underlying neural network or is individually encoded in the voxel map”and,
the training at least the first model (C) further comprises jointly training a second model and a third model with the first model, wherein the second model has a second plurality of parameters and the third model has a third plurality of parameters, the training further using, for each corresponding training compound in the plurality of training compounds, at least:(iii) the corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer as input to the second model, against the corresponding second positive interaction score of the corresponding training compound with respect to the target polymer, (Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”. Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”) and
v) the corresponding positive score for the corresponding positive pose of the
corresponding training compound with respect to the target polymer, the output of the
first model and the output of the second model upon input of the corresponding positive
score for the corresponding positive pose of the corresponding training compound as joint input to the third model, against the corresponding positive activity score of the
corresponding training compound, (Heifets, Figure 2E, steps 282,284,286,288, 290, [0009] “The at least one program further comprises instructions for obtaining a plurality of scores from the scorer, where each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors (or a corresponding voxel map) into the input layer. The plurality of scores is used to determine a classification of the test object. In some embodiments, a weighted average of the plurality of scores is used to determine a classification of the test object”) and
Heifets is silent on a corresponding second negative interaction score for the corresponding negative pose of the corresponding training compound
(iv) the corresponding negative score for the corresponding negative pose of the
corresponding training compound with respect to the target polymer as input to the
second model, against the corresponding second negative interaction score of the
corresponding training compound with respect to the target polymer, thereby adjusting
the second plurality of parameters,
(vi) the corresponding negative score for the corresponding negative pose of the
corresponding training compound with respect to the target polymer and the output of the first model and the output of the second model upon input of the corresponding negative score for the corresponding negative pose of the corresponding training compound as joint input to the third model, against the corresponding negative activity score of the corresponding training compound, thereby adjusting the third plurality of parameters of the third model, wherein an output of the third model provides the characterization of the interaction between the test compound and the target polymer.
However, Ragoza teaches a corresponding second negative interaction score for the corresponding negative pose of the corresponding training compound(iv) the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the second model, against the corresponding second negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the second plurality of parameters, (Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
(vi) the corresponding negative score for the corresponding negative pose of the
corresponding training compound with respect to the target polymer and the output of the first model and the output of the second model upon input of the corresponding negative score for the corresponding negative pose of the corresponding training compound as joint input to the third model, against the corresponding negative activity score of the corresponding training compound, thereby adjusting the third plurality parameters of the third model, wherein an output of the third model provides the characterization of the interaction between the test compound and the target polymer.(Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 21, combination of Heifets and Ragoza teaches the computer system of claim 20,
Heifets further teaches wherein the second model is a second fully connected neural network, and the third model is a third fully connected neural network. (Heifets, Figure 2E, [0131] In some embodiments, the scorer 30 comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully connected layers feeds into the evaluation layer (272)” NOTE: fully connected plurality of layers reads on second, third fully connected neural network.).
Regarding Claim 22, combination of Heifets and Ragoza teaches the computer system of claim 20,
Heifets further teaches wherein the corresponding positive activity score is a first binary activity score and the corresponding negative activity score is a second binary activity score. (Heifets, Figure 2A, 238 “The plurality of channels includes a first channel that is a binary categorical variable for presence of an atom in the test object and second channel that is a binary categorical variable for presence of an atom in the target object”).
Regarding Claim 23, combination of Heifets and Ragoza teaches the computer system of claim 22,
Heifets further teaches wherein the corresponding first binary activity score is assigned a value of 1 based on a measured activity of the corresponding compound against the target polymer, and the corresponding second binary activity score is assigned a value of 0. (Heifets, [0086], When a given atom type is found in the three-dimensional grid element corresponding to a given voxel, the channel for
that atom type within the given voxel is assigned a first value of the binary categorical variable, such as "1", and when the atom type is not found in the three-dimensional grid element corresponding to the given voxel, the channel for that atom type is assigned a second value of the binary categorical variable, such as "0" within the given voxel”).
Regarding Claim 24, combination of Heifets and Ragoza teaches the computer system of claim 20,
Heifets further teaches wherein the training of the first model is a first regression task in which the first plurality of parameters is adjusted by back-propagation through a first associated loss function, (Heifets, [0008] a multiple additive regression tree, a clustering algorithm, principal component analysis, a nearest neighbor analysis (…) or ensembles thereof. [0150] Training the Predictive Model. In some embodiments, where a deep neural network is implemented (e.g.,the convolutional neural network 24), the convolutional assessment module 20 is configured to train the network 24 to receive the geometric data input and to output a prediction (probability) of whether or not a given test object binds to a target object. t [0158] In an embodiment, the neural network 24 may optionally, where training data is labeled (e.g., with the binding data 68), tune the weights within the network 24 to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations. Various methods may be used to minimize error function, such as gradient descent methods, which may include, but are not limited to, log-loss, sum of squares error, hinge-loss methods”. [0151] For instance, the filter weights of respective filters in the convolutional layers 28 of the network are adjusted in such back-propagation”),
the training of the second model is a second regression task in which the second plurality of parameters is adjusted by back-propagation through a second associated loss function, and the training of the third model is a classification task in which the third plurality of parameters is adjusted by back-propagation through a third associated loss function. (Heifets, see [0150] Training the Predictive Model. [0158], [0151] as above. NOTE: fully connected plurality of layers reads on second, third fully connected neural network.).
Regarding Claim 25, combination of Heifets and Ragoza teaches the computer system of claim 24,
Heifets further teaches, wherein the corresponding first positive interaction score and each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer. (Heifets, [0090] In some embodiments, a structural protein-ligand interaction fingerprint (SPLIF) score is generated for each pose of a given test object (or training object) to a target
object and this SPLIF score is used as additional input into the underlying neural network or is individually encoded in the voxel map.).
Heifets is silent on the corresponding first negative interaction score each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer
However, Ragoza teaches the corresponding first negative interaction score each represent a binding coefficient or an in silico pose quality score of the corresponding training compound to the target polymer ;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 26, combination of Heifets and Ragoza teaches the computer system of claim 24,
Heifets further teaches wherein the first associated loss function is a mean squared error loss function, a mean absolute error loss function, a Huber loss function, a Log-Cosh loss function, or a quantile loss function, the second associated loss function is a mean squared error loss function, a mean absolute error loss function, a Huber loss function, a Log-Cosh loss function, or a quantile loss function, and the third associated loss function is a binary cross entropy loss function, a hinge loss function, or a squared hinged loss function. (Heifets, [0008] a multiple additive regression tree, a clustering algorithm, principal component analysis, a nearest neighbor analysis (…) or ensembles thereof. [0150] Training the Predictive Model. In some embodiments, where a deep neural network is implemented (e.g.,the convolutional neural network 24), the convolutional assessment module 20 is configured to train the network 24 to receive the geometric data input and to output a prediction (probability) of whether or not a given test object binds to a target object. t [0158] In an embodiment, the neural network 24 may optionally, where training data is labeled (e.g., with the binding data 68), tune the weights within the network 24 to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations. Various methods may be used to minimize error function, such as gradient descent methods, which may include, but are not limited to, log-loss, sum of squares error, hinge-loss methods”).
Regarding Claim 27, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the polymer is a protein, a polypeptide, a polynucleic acid, a polyribonucleic acid, a polysaccharide, or an assembly of any combination thereof. (Heifets, Figure 2A, 206 The polymer is a protein, a polypeptide, a polynucleic acid, a polyribonucleic acid, a polysaccharide, or an assembly of any combination thereof”).
Regarding Claim 28, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the plurality of atomic coordinates is a set of three-dimensional coordinates for a crystal structure of the target polymer resolved at a resolution of 2.5 A or better or a resolution of 3 .3 A or better. (Heifets, Figure 2A, 208,210, “The target object is a polymer and the spatial coordinates are a set of three-dimensional coordinates {x lr- 1, ... , xN} for a crystal structure of the polymer resolved at a resolution of 2.5 A or better. a resolution of 3.3 A or better”).
Regarding Claim 29, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the plurality of atomic coordinates for the target polymer comprises an ensemble of three- dimensional coordinates for the target polymer determined by nuclear magnetic resonance, neutron diffraction, or cryo-electron microscopy. (Heifets, Figure 2A,212, The target object is a polymer and the spatial coordinates are an ensemble of three-dimensional coordinates of the polymer determined by nuclear magnetic resonance, neutron diffraction, or cryo-electron
Microscopy”).
Regarding Claim 30, combination of Heifets and Ragoza teaches the computer system of claim 1,
Heifets further teaches wherein the characterization of the interaction between the test compound and the target polymer is a binary score, (Heifets, Figure 2A, 238 “The plurality of channels includes a first channel that is a binary categorical variable for presence of an atom in the test object and second channel that is a binary categorical variable for presence of an atom in the target object”).
wherein a first value for the binary score represents an ICso, ECso, Kd, KI, or pKI for the test compound with respect to the target polymer that is above a first threshold, (Heifets, Figure 2E, The first classification is an IC50, EC50, Kd or Kl for the test object with respect to the target object that is above a first binding value (e.g., one vmicromolar, ten micromolar, and the second classification is an IC50,EC50, or Kl for the test object with respect to the target object that is below the first binding value and a second value for the binary score represents an ICso, ECso, Kd, KI, or pKI for the test compound with respect to the target polymer that is below the first threshold (Heifets, Figure 2E, 290, Each respective classification in the plurality of classifications is an IC50, EC50, Kd, or Kl range (e.g., between one micromolar and ten micromolar. between one nanomolar and 100 nanomolar) for the test object with respect to the target object”)
Regarding Claim 31, Heifets teaches the computer system of claim 1,
Heifets further teaches wherein each training compound in the training dataset satisfies two or more rules, three or more rules, or all four rules of the Lipinski's rule of Five: (i) not more than five hydrogen bond donors, (ii) not more than ten hydrogen bond acceptors, (iii) a molecular weight under 500 Daltons, and (iv) a LogP under 5.(Heifets, Figure 2a, [0069] “In some embodiments, test objects 72 and training objects 66 are organic compounds that satisfy two or more mies, three or more rules, or all four rules of the Lipinski's Rule of Five: (i) not more than five hydrogen bond donors (e.g., OH and NH groups), (ii) not more than ten hydrogen bond acceptors ( e.g. N and 0), (iii) a molecular weight under 500 Daltons, and (iv) a Log P lillder 5. The "Rule of Five" is so called because three of the four criteria involve the number five”).
Regarding Claim 32, Heifets teaches the computer system of claim 1,
Heifets further teaches wherein each training compound in the training dataset is an organic compound having a molecular weight of less than 4000 Daltons (Heifets, [0070] “In some embodiments, a test object 72 or training object 66 satisfies one or more criteria in addition to Lipinski 's Rule of Five; test object 72 or training object 66 is any organic compound having a molecular weight of less than 2000 Daltons,of less than 4000 Daltons”).
Regarding Claim 33, Heifets teaches the computer system of claim 1,
Heifets further teaches wherein the corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer is obtained from a convolutional neural network upon inputting the corresponding positive pose of the corresponding training compound with respect to the target polymer into the convolutional neural network, (Heifets, Figure 2a, [0073] In some embodiments, the test object 72 or training object 66 is docked onto the target object 58 a plurality of times to form a plurality of poses. In some embodiments, the test object 72 or training object 66 is docked onto the target object 58 twice, three times, four times, five or more times, ten or more times, fifty or more times, 100 or more times, or a 1000 or more times (218). Each such docking represents a different pose of the test object 72 or training object 66 docked onto the target object 58”) and
Heifets is silent on the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer is obtained from the convolutional neural network upon inputting the corresponding negative pose of the corresponding training compound with respect to the target polymer into the convolutional neural network.
However, Ragoza teaches the corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer is obtained from the convolutional neural network upon inputting the corresponding negative pose of the corresponding training compound with respect to the target polymer into the convolutional neural network.;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 34, Heifets teaches the computer system of claim 33,
Heifets further teaches, wherein the convolutional neural network is a graph convolutional neural network, an equivariant neural network, or a message passing neural network. (Heifets, Figure 1,24, Convolutional neural network).
Regarding Claim 35, Heifets teaches,
A method for characterizing an interaction between a test compound and a target
polymer, (Heifets, Figure 1) the method comprising:
at a computer system comprising a memory:
(A) obtaining a plurality of atomic coordinates for the target polymer(Heifets, Figure 1, [0003] “the prediction the affinity of test
objects to target objects”. [0007] “target object (e.g., a polymer)”) the computer system comprising:
one or more processors (Heifets, figure 2a, [0048] the computer system 100 comprises a general processor 74”); and
memory addressable by the one or more processors, the memory storing at least one program for execution by the one or more processors, the at least one program comprising instructions (Heifetz, figure 2a, [0048] “the computer system 100 comprises a general processor 74. general memory 90 / 92 addressable by the general processing unit, the general memory storing at least one program 56 for execution by the at least one general processor”); for:
(A) obtaining a plurality of atomic coordinates for the target polymer, wherein the plurality of atomic coordinates comprises atomic coordinates for; (Heifets, figure 2a, step 208, [0007] “The at least one program comprises instructions for obtaining a set of three-dimensional coordinates {xi, ... , xN} for all or a portion of a target object ( e.g., a polymer having an active site). Each respective x, in {xi, ... , xN} is a three dimensional coordinate for an atom in a plurality of atoms in the target object”). at least 400 atoms (Heifets teaches in [0071] “the systems and methods of the present disclosure have no limitation on the size of the test objects 72 or training objects 66. For instance, in some embodiments, such objects are large polymer, such as antibodies [0061]” In some embodiments the target object is a polymer and there are ten or more, twenty or more, thirty or more, fifty or more, one hundred or more, between one hundred and one thousand, or less than 500 residues in the polymer”. a polymer with 100 or more residues means the chain is composed of 100 or more repeating units connected together. Each polymer unit contains multiple atoms depends on the length and types of polymers. Therefore, for example a polymer is a protein polymer on average contains 12-15 atoms and with 100 residues total atoms will be 1200, which is more than 400 atoms. This is known in the art.)
(B) obtaining a training dataset comprising a respective electronic description of each training compound in a plurality of training compounds, wherein the plurality of training compounds comprises at least 100 compounds, (Heifets, FIG. 1, 92, Object training library, Training object 66-1, Figure 2A, In step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”. In step 218 “The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 or more poses”. “ plurality of poses” reads on the electronic description of each training compound. Also see [0073])
each respective electronic description comprising:
(i) a corresponding positive pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first positive interaction score (Heifets, Figure 2a, 212,” The target object is a polymer and the spatial coordinates are an ensemble of three-dimensional coordinates of the polymer”. Figure 2A, step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”.[ 0139] Obtaining a Plurality of Scores from the Scorer (276) and Using a Score from the Convolutional Neural Network to Characterize the Test Object (278)”), and
(C) training at least a first model, wherein the first model has a first plurality of parameters (Heitfets, Figure 2A,step 214-216, “ Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses. The target object 58 is a polymer with an active site, the test object is a chemical compound, and the docking comprises docking the test object into the active site of the polymer” , and wherein the first plurality of parameters comprises more than 400 parameters, the training using, for each corresponding training compound in the plurality of training compounds, at least (Heifets, [0010] (i) sampling the test object, in a respective pose in the plurality of different poses, and a portion of the target object that is contact with the test object in the respective pose ( e.g., the active site of the target object in instance where the target object is a polymer) on a three-dimensional basis (e.g., grid basis) thereby forming a corresponding three dimensional space filling uniform honeycomb comprising a corresponding plurality of three-dimensional space filling polyhedral cells. For each respective three-dimensional cell in the corresponding plurality of three-dimensional cells, a voxel in the respective voxel map is populated based upon one or more properties (e.g., chemical properties such as distance, angle, atom type, charge and polarization, and surrounding stabilizing or destabilizing environmental factors) of the respective three-dimensional cell”. Figure 2A, step 214, “ The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 more poses.”. Therefore, for each respective pose corresponding plurality of parameters, therefore for 100 or more poses with multiple parameters accounted for more than 400 parameters.)
(i) a corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first positive interaction score of the corresponding training compound with respect to the target polymer (Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”. Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”) and
wherein at least an output of the first model is used, at least in part, to provide the characterization of the interaction between the test compound and the target polymer. (Heifets, Figure 2E, steps 282,284,286,288, 290, [0009] “The at least one program further comprises instructions for obtaining a plurality of scores from the scorer, where each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors (or a corresponding voxel map) into the input layer. The plurality of scores is used to determine a classification of the test object. In some embodiments, a weighted average of the plurality of scores is used to determine a classification of the test object”)
Heifets teaches incorrect poses when meaning that such poses do not represent true interactions or binding energy between the test object 72 (or training object 66) and the target object 58. The wrong interactions are example of non-binding interactions. see Heifets, [0073]”. However, Heifets is silent on negative score determination.
Heifets is silent on (ii) a corresponding negative pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first negative interaction score;
(ii) a corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the first plurality of parameters,
However, Ragoza teaches ii) a corresponding negative pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first negative interaction score;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
(ii) a corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the first plurality of parameters, (Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Regarding Claim 36, Heifets teaches,
A non-transitory computer readable storage medium, wherein the non-transitory
computer readable storage medium stores instructions, which when executed by a computer system, cause the computer system to perform a method for characterizing an interaction between a test compound and a target polymer, (Heifetz, figure 2a, [0048] “the computer system 100 comprises a general processor 74. general memory 90 / 92 addressable by the general processing unit, the general memory storing at least one program 56 for execution by the at least one general processor” Heifets, figure 2a, [0048] the computer system 100 comprises a general processor) the method comprising:
(A) obtaining a plurality of atomic coordinates for the target polymer (Heifets, 74”); and Figure 1, [0003] “the prediction the affinity of test objects to target objects”. [0007] “target object (e.g., a polymer)”) the computer system comprising:
(A) obtaining a plurality of atomic coordinates for the target polymer, wherein the plurality of atomic coordinates comprises atomic coordinates for; (Heifets, figure 2a, step 208, [0007] “The at least one program comprises instructions for obtaining a set of three-dimensional coordinates {xi, ... , xN} for all or a portion of a target object ( e.g., a polymer having an active site). Each respective x, in {xi, ... , xN} is a three dimensional coordinate for an atom in a plurality of atoms in the target object”). at least 400 atoms (Heifets teaches in [0071] “the systems and methods of the present disclosure have no limitation on the size of the test objects 72 or training objects 66. For instance, in some embodiments, such objects are large polymer, such as antibodies [0061]” In some embodiments the target object is a polymer and there are ten or more, twenty or more, thirty or more, fifty or more, one hundred or more, between one hundred and one thousand, or less than 500 residues in the polymer”. a polymer with 100 or more residues means the chain is composed of 100 or more repeating units connected together. Each polymer unit contains multiple atoms depends on the length and types of polymers. Therefore, for example a polymer is a protein polymer on average contains 12-15 atoms and with 100 residues total atoms will be 1200, which is more than 400 atoms. This is known in the art.)
(B) obtaining a training dataset comprising a respective electronic description of each training compound in a plurality of training compounds, wherein the plurality of training compounds comprises at least 100 compounds, (Heifets, FIG. 1, 92, Object training library, Training object 66-1, Figure 2A, In step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”. In step 218 “The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 or more poses”. “ plurality of poses” reads on the electronic description of each training compound. Also see [0073])
each respective electronic description comprising:
(i) a corresponding positive pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first positive interaction score (Heifets, Figure 2a, 212,” The target object is a polymer and the spatial coordinates are an ensemble of three-dimensional coordinates of the polymer”. Figure 2A, step 214” Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses”.[ 0139] Obtaining a Plurality of Scores from the Scorer (276) and Using a Score from the Convolutional Neural Network to Characterize the Test Object (278)”), and
(C) training at least a first model, wherein the first model has a first plurality of parameters (Heitfets, Figure 2A,step 214-216, “ Model the test objects 72 and/or training objects 66 with the target objects 58 in each pose of a plurality of different poses. The target object 58 is a polymer with an active site, the test object is a chemical compound, and the docking comprises docking the test object into the active site of the polymer” , and wherein the first plurality of parameters comprises more than 400 parameters, the training using, for each corresponding training compound in the plurality of training compounds, at least (Heifets, [0010] (i) sampling the test object, in a respective pose in the plurality of different poses, and a portion of the target object that is contact with the test object in the respective pose ( e.g., the active site of the target object in instance where the target object is a polymer) on a three-dimensional basis (e.g., grid basis) thereby forming a corresponding three dimensional space filling uniform honeycomb comprising a corresponding plurality of three-dimensional space filling polyhedral cells. For each respective three-dimensional cell in the corresponding plurality of three-dimensional cells, a voxel in the respective voxel map is populated based upon one or more properties (e.g., chemical properties such as distance, angle, atom type, charge and polarization, and surrounding stabilizing or destabilizing environmental factors) of the respective three-dimensional cell”. Figure 2A, step 214, “ The plurality of different poses comprises 2 or more poses, 10 or more poses, 100 or more poses, or 1000 more poses.”. Therefore, for each respective pose corresponding plurality of parameters, therefore for 100 or more poses with multiple parameters accounted for more than 400 parameters.)
(i) a corresponding positive score for the corresponding positive pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first positive interaction score of the corresponding training compound with respect to the target polymer (Heifets, Figure 2D; 266, “each respective convolutional layer, other than the final convolutional layer, feeds intermediate values, as a respective second function of (i) the different set of weights associated with the respective convolutional layer and (ii) input values received by the respective convolutional layer, into another convolutional layer in the plurality of convolutional layers, wherein the second function is computed using the graphical processing unit. The scorer comprises a plurality of fully-connected layers and an evaluation layer where a fully-connected layer in the plurality of fully-connected layers feeds into the evaluation layer”. Figure 2E, 276, Obtaining a plurality of scores from the scorer, wherein each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors into the input layer”) and
wherein at least an output of the first model is used, at least in part, to provide the characterization of the interaction between the test compound and the target polymer. (Heifets, Figure 2E, steps 282,284,286,288, 290, [0009] “The at least one program further comprises instructions for obtaining a plurality of scores from the scorer, where each score in the plurality of scores corresponds to the input of a vector in the plurality of vectors (or a corresponding voxel map) into the input layer. The plurality of scores is used to determine a classification of the test object. In some embodiments, a weighted average of the plurality of scores is used to determine a classification of the test object”)
Heifets teaches incorrect poses when meaning that such poses do not represent true interactions or binding energy between the test object 72 (or training object 66) and the target object 58. The wrong interactions are example of non-binding interactions. see Heifets, [0073]”. However, Heifets is silent on negative score determination.
Heifets is silent on (ii) a corresponding negative pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first negative interaction score;
(ii) a corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the first plurality of parameters,
However, Ragoza teaches ii) a corresponding negative pose of the corresponding training compound with respect to the plurality of atomic spatial coordinates coupled with a corresponding first negative interaction score;(Ragoza, page 5, top paragraph, “We utilize two training sets focused on two different goals: pose prediction and virtual screening. In all cases we generate ligand poses for actives and decoys using docking with smina1 and the AutoDock Vina scoring function.7 We use docked poses, even for active compounds with a known crystal structure, because (1) these are the types of poses the model will ultimately have to score and (2) to avoid the model simply learning to distinguish between docked poses and crystal structures (which were likely optimized with different force fields”)
(ii) a corresponding negative score for the corresponding negative pose of the corresponding training compound with respect to the target polymer as input to the first model, against the corresponding first negative interaction score of the corresponding training compound with respect to the target polymer, thereby adjusting the first plurality of parameters, (Ragoza, page 4, third paragraph, Here, we describe the evelopment of a CNN model for protein-ligand scoring that is trained to classify compound poses as binders or non-binders using a 3D grid representation of proteinligand
structures generated through docking. We show that our CNN scoring method outperforms the AutoDock Vina7 scoring function that is used to generate the poses both when selecting poses for pose prediction and for virtual screening tasks. We also illustrate how our CNN score can be decomposed into individual atomic contributions to generate informative visualizations”).
It would have been obvious to a person having ordinary skill in the art before the effective filing date to modify Heifets’s method to incorporate a method of estimating negative poses and negative score for non-binding interactions between compound and polymers along with machine learning model as taught by Ragoza and obtain an accurate compound classification based on binders and non-binders scoring. (Ragoza, Abstract). It would have been obvious to a person of ordinary skill to include the well-known negative pose and negative score calculation method along with the other machine learning network, in order to yield the predicted results of generating compound and polymer interaction characteristics and classifying with higher accuracy (KSR).
Conclusion
Citation of Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Deng et al. (US 20070020642 A1) describes.”a method for representing and analyzing 3D target molecule-ligand intermolecular interactions. The method generates structural interaction fingerprints (SIFts) that convert three-dimensional structural interaction information into linear information strings that contains a plurality of information blocks; each of which in turn containing a plurality of information units. By assigning to each information unit a calculated value to represent the characteristic of a set of intermolecular interactions occurring at each selected position (i.e., a position on the target molecule at which intermolecular interaction occurs), a SIFt of the target molecule-ligand complex is constructed” (abstract)
Prakash et al. (US2012116742 (A1) The invention provides “Method and apparatus for the efficient computation of values for affinity functions for two or more molecular subsets of a molecular configuration, are provided. Either one or both of molecular subsets may be selected from a molecule library. Affinity engines can compute the affinity values, and can be synchronized in order to maximize utilization of processing power available in the affinity engines. A data path allocator can apportion molecular descriptor data to each affinity engine as one or more data blocks according to a data path schedule. Also, new configurations may be generated from one or more input configurations, computation of a plurality of affinity values for a plurality of configurations, and subsequent selection of processed configurations for further analysis (abstract).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DILARA SULTANA whose telephone number is (571)272-3861. The examiner can normally be reached Mon-Fri, 9 AM-5:30 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s
supervisor, EMAN ALKAFAWI can be reached on (571) 272-4448. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/DILARA SULTANA/Examiner, Art Unit 2858 July 23, 2026
/EMAN A ALKAFAWI/Supervisory Patent Examiner, Art Unit 2858 7/27/2026
Prosecution Insights