Abstract: A system and method for pharmacophore-conditioned generation of molecules. The system and method modifies a conditional variational autoencoder (CVAE) such that the latent space in generation of a molecule is not conditioned on the pharmacophore space of the molecule. This allows for generation of pharmacophore descriptors independently from the conditional on which CVAE has been trained, removing a substantial impediment to the use of CVAEs for exploration of pharmacophore descriptors of a molecule.

Abstract: A system and method comprising a transmoler that identifies common substructures of a given 3D conformer and predicts its structural information. First, based on contrastive learning, substructure embeddings are learned in an unsupervised manner. Secondly, a novel oriented 3D object regressor predicts the dimensions and directions of each substructure in a conformer as well as its fingerprint embedding which are used to create differentiable junction tree molecular graphs. Lastly, using the junction tree graphs, molecular representations such as DeepSMILES are generated which represent new and novel molecules. The system may also generate conformers directly from a pocket. A pocket may be input to the model and the model learns to generate structures which can fit that pocket by conditioning the generative system. Furthermore, structure-based contrastive embeddings generated for transmoler can be recycled in structure-based generative modelling.

Abstract: A system and method for de novo drug discovery using machine learning algorithms. In a preferred embodiment, de novo drug discovery is performed via data enrichment and interpolation/perturbation of molecule models within the latent space, wherein molecules with certain characteristics can be generated and tested in relation to one or more targeted receptors. Filtering methods may be used to determine active novel molecules by filtering out non-active molecules and contain activity predictors to better navigate the molecule-receptor domain. The system may comprise neural networks trained to reconstruct known ligand-receptors pairs and from the reconstruction model interpolate and perturb the model such that novel and unique molecules are discovered. A second preferred embodiment trains a variational autoencoder coupled with a bioactivity model to predict molecules exhibiting a range of desired properties.

Abstract: A system and method for optimizing the latent space in generative machine learning models, and applications of the optimizations for use in the de novo generation of molecules for both ligand-based and pocket-based generation. The ligand-based optimizations comprise a tunable reward system based on a multi-property model and further define new measurable metrics: molecular novelty and uniqueness. The pocket-based optimizations comprise an initial multi-property optimization followed up by either a seed-based optimization or a relaxed-based optimization.

Abstract: A system and method for pharmacophore-conditioned generation of molecules. The system and method modifies a conditional variational autoencoder (CVAE) such that the latent space in generation of a molecule is not conditioned on the pharmacophore space of the molecule. This allows for generation of pharmacophore descriptors independently from the conditional on which CVAE has been trained, removing a substantial impediment to the use of CVAEs for exploration of pharmacophore descriptors of a molecule.

Abstract: A system and method for de novo drug discovery using machine learning algorithms. In a preferred embodiment, de novo drug discovery is performed via data enrichment and interpolation/perturbation of molecule models within the latent space, wherein molecules with certain characteristics can be generated and tested in relation to one or more targeted receptors. Filtering methods may be used to determine active novel molecules by filtering out non-active molecules and contain activity predictors to better navigate the molecule-receptor domain. The system may comprise neural networks trained to reconstruct known ligand-receptors pairs and from the reconstruction model interpolate and perturb the model such that novel and unique molecules are discovered. A second preferred embodiment trains a variational autoencoder coupled with a bioactivity model to predict molecules exhibiting a range of desired properties.

Abstract: A system and method for optimizing the latent space in generative machine learning models, and applications of the optimizations for use in the de novo generation of molecules for both ligand-based and pocket-based generation. The ligand-based optimizations comprise a tunable reward system based on a multi-property model and further define new measurable metrics: molecular novelty and uniqueness. The pocket-based optimizations comprise an initial multi-property optimization followed up by either a seed-based optimization or a relaxed-based optimization.

Abstract: A system and method for accelerating the calculations of free energy differences by automating FEP-path-decision-making and replacing the standard series of alchemical interpolations typically created by molecular dynamic (MD) simulations with voxelated interpolated states. A novel machine learning approach comprising a restricted variational autoencoder (ResVAE) is used which can reduce the computational-cost associated with interpolations by restricting the dimensions of a molecular latent space. The ResVAE generates a model based on flow-based transformations of a 3D-VAE latent point that is trained to maximize the log-likelihood of MD samples which enables the model to compute transformations more efficiently between molecules and also handle deletions of atoms more efficiently during iterative FEP calculation steps.

Abstract: A system and method for optimizing the latent space in generative machine learning models, and applications of the optimizations for use in the de novo generation of molecules for both ligand-based and pocket-based generation. The ligand-based optimizations comprise a tunable reward system based on a multi-property model and further define new measurable metrics: molecular novelty and uniqueness. The pocket-based optimizations comprise an initial multi-property optimization followed up by either a seed-based optimization or a relaxed-based optimization.

Abstract: A system and method for accelerating the calculations of free energy differences by automating FEP-path-decision-making and replacing the standard series of alchemical interpolations typically created by molecular dynamic (MD) simulations with voxelated interpolated states. A novel machine learning approach comprising a restricted variational autoencoder (ResVAE) is used which can reduce the computational-cost associated with interpolations by restricting the dimensions of a molecular latent space. The ResVAE generates a model based on flow-based transformations of a 3D-VAE latent point that is trained to maximize the log-likelihood of MD samples which enables the model to compute transformations more efficiently between molecules and also handle deletions of atoms more efficiently during iterative FEP calculation steps.

Abstract: A system and method comprising a transmoler that identifies common substructures of a given 3D conformer and predicts its structural information. First, based on contrastive learning, substructure embeddings are learned in an unsupervised manner. Secondly, a novel oriented 3D object regressor predicts the dimensions and directions of each substructure in a conformer as well as its fingerprint embedding which are used to create differentiable junction tree molecular graphs. Lastly, using the junction tree graphs, molecular representations such as DeepSMILES are generated which represent new and novel molecules. The system may also generate conformers directly from a pocket. A pocket may be input to the model and the model learns to generate structures which can fit that pocket by conditioning the generative system. Furthermore, structure-based contrastive embeddings generated for transmoler can be recycled in structure-based generative modelling.

Abstract: A system and method that takes in a data set comprising molecular structure data and properties of interest, e.g., ADMET, EC50, IC50, etc., and determines the substructures that cause or do not cause the property of interest. The substructures may then be used to filter out potentially harmful new proposed/generated molecules or create a new data set of known active/inactive substructures of a property of interest that may fulfill other obligations. The system comprises a substructure extraction module which further comprises a scaffold extraction module and a comparison module. A scaffold extraction module clusters, searches, and extracts substructures in question while a comparison module compares the bioactivity of each molecule with and without each substructure in question to determine the substructures effect on the property of interest.

Abstract: A system and method for de novo drug discovery using machine learning algorithms. In a preferred embodiment, de novo drug discovery is performed via data enrichment and interpolation/perturbation of molecule models within the latent space, wherein molecules with certain characteristics can be generated and tested in relation to one or more targeted receptors. Filtering methods may be used to determine active novel molecules by filtering out non-active molecules and contain activity predictors to better navigate the molecule-receptor domain. The system may comprise neural networks trained to reconstruct known ligand-receptors pairs and from the reconstruction model interpolate and perturb the model such that novel and unique molecules are discovered. A second preferred embodiment trains a variational autoencoder coupled with a bioactivity model to predict molecules exhibiting a range of desired properties.

Abstract: A system and method that takes in a data set comprising molecular structure data and properties of interest, e.g., ADMET, EC50, IC50, etc., and determines the substructures that cause or do not cause the property of interest. The substructures may then be used to filter out potentially harmful new proposed/generated molecules or create a new data set of known active/inactive substructures of a property of interest that may fulfill other obligations. The system comprises a substructure extraction module which further comprises a scaffold extraction module and a comparison module. A scaffold extraction module clusters, searches, and extracts substructures in question while a comparison module compares the bioactivity of each molecule with and without each substructure in question to determine the substructures effect on the property of interest.

Abstract: A system and method for automated retrosynthesis which can reliably identify valid and practical precursors and reaction pathways. The methodology involves a k-beam recursive process wherein at each stage of recursion, retrosynthesis is performed using a library of molecule disconnection rules to identify possible precursor sets, validation of the top k precursor sets is performed using a transformer-based forward reaction prediction scoring system, the best candidate of the top k precursor sets is selected, and a database is searched to determine whether the precursors are commercially available. The recursion process is repeated until a valid chain of chemical reactions is found wherein all precursors necessary to synthesize the target molecule are found to be commercially available.

Abstract: A system and method for automated retrosynthesis which can reliably identify valid and practical precursors and reaction pathways. The methodology involves a k-beam recursive process wherein at each stage of recursion, retrosynthesis is performed using a library of molecule disconnection rules to identify possible precursor sets, validation of the top k precursor sets is performed using a transformer-based forward reaction prediction scoring system, the best candidate of the top k precursor sets is selected, and a database is searched to determine whether the precursors are commercially available. The recursion process is repeated until a valid chain of chemical reactions is found wherein all precursors necessary to synthesize the target molecule are found to be commercially available.

Abstract: A system and method that produces an accurate probability distribution representative of a target molecule that may be used in pharmacokinetics and analogous applications. A generator is seeded from a variational autoencoder during training and is then used after training in series with a second variational autoencoder to produce the probability distributions from molecular tensors.

Abstract: A system and method that predicts whether a given protein-ligand pair is active or inactive, the ground-truth protein-ligand complex crystalline-structure similarity, and an associated bioactivity value. The system and method further produce 3-D visualizations of previously unknown protein-ligand pairs that show directly the importance assigned to protein-ligand interactions, the positive/negative-ness of the saliencies, and magnitude. Furthermore, the system and method make enhancements in the art by accurately predicting protein-ligand pair bioactivity from decoupled models, removing the need for docking simulations, as well as restricting attention of the machine learning between protein and ligand atoms only.

Abstract: A system and method that predicts whether a given protein-ligand pair is active or inactive and outputs a pose score classifying the propriety of the pose. A 3D bioactivity platform comprising a 3D bioactivity module and data platform scrapes empirical lab-based data that a docking simulator uses to generate a dataset from which a 3D-CNN model is trained. The model then may receive new protein-ligand pairs and determine a classification for the bioactivity and pose propriety of that protein-ligand pair. Furthermore, gradients relating to the binding affinity in the 3D model of the molecule may be used to generate profiles from which new protein targets may be determined.

Abstract: A system and method for computationally tractable prediction of protein-ligand interactions and their bioactivity. According to an embodiment, the system and method comprise two machine learning processing streams and concatenating their outputs. One of the machine learning streams is trained using information about ligands and their bioactivity interactions with proteins. The other machine learning stream is trained using information about proteins and their bioactivity interactions with ligands. After the machine learning algorithms for each stream have been trained, they can be used to predict the bioactivity of a given protein-ligand pair by inputting a specified ligand into the ligand processing stream and a specified protein into the protein processing stream. The machine learning algorithms of each stream predict possible protein-ligand bioactivity interactions based on the training data.