Our research group focuses upon the application and development of new computational tools in drug discovery research, through the application of different Artificial Intelligence (AI) approaches.
Drug Discovery Projects
We currently work on several drug discovery projects. We collaborate with experimental scientists across the country with the goal of developing chemical modulators for the treatment of multiple diseases. We bring chemical entities to biology using computational tools like docking, pharmacophore modeling, molecular dynamics, ADME predictions, and homology modeling. We are particularly interested in studying membrane protein such as G-Protein Coupled receptors. We employ enhanced molecular dynamics simulations such as accelerated Molecular dynamics, Metadynamics , and Free Energy Perturbations.
We currently work on several drug discovery projects. We collaborate with experimental scientists across the country with the goal of developing chemical modulators for the treatment of multiple diseases. We bring chemical entities to biology using computational tools like docking, pharmacophore modeling, molecular dynamics, ADME predictions, and homology modeling. We are particularly interested in studying membrane protein such as G-Protein Coupled receptors. We employ enhanced molecular dynamics simulations such as accelerated Molecular dynamics, Metadynamics , and Free Energy Perturbations.
DLSCORE: A Deep Learning Model for Predicting Protein-Ligand Binding Affinities
We have developed a Deep Learning model that can accurately predict the binding affinities between a protein and a ligand. We employed fully-connected neural networks (FCNN). The networks in the model consist of multiple hidden layers with a different number of neurons. Each hidden layer has a weight matrix (W) with a dimension ruled by the input size and the number of neurons in that layer. During our testing stage we obtained a Pearson R2 of 0.82, an RMSE of 1.15, and other significant results, suggesting our model is appropriate for binding affinity prediction between a protein and a ligand. For more information, take a look at our paper at:
We have developed a Deep Learning model that can accurately predict the binding affinities between a protein and a ligand. We employed fully-connected neural networks (FCNN). The networks in the model consist of multiple hidden layers with a different number of neurons. Each hidden layer has a weight matrix (W) with a dimension ruled by the input size and the number of neurons in that layer. During our testing stage we obtained a Pearson R2 of 0.82, an RMSE of 1.15, and other significant results, suggesting our model is appropriate for binding affinity prediction between a protein and a ligand. For more information, take a look at our paper at:
Open Drug Metabolism & Pharmacokinetics (OpenDMPK)
We are currently developing a benchmark where the user is allowed to choose among different Machine Learning models in order to predict specific pharmacokinetics (PK) and drug metabolism (DM) properties from a wide variety of ligands. These predictions may fit into a regression or classification mode, for instance, the user may want to predict the solubility of a ligand (regression problem) or whether it crosses the blood brain barrier (BBB) or not (classification problem). For this, we are providing different models that include Support Vector Machine (SVM), Random Forests (RF), Multi Linear Regression, among others. Predicting this properties will help the user to speed up the process for the pre-clinical trials in regards of drug discovery.
We are currently developing a benchmark where the user is allowed to choose among different Machine Learning models in order to predict specific pharmacokinetics (PK) and drug metabolism (DM) properties from a wide variety of ligands. These predictions may fit into a regression or classification mode, for instance, the user may want to predict the solubility of a ligand (regression problem) or whether it crosses the blood brain barrier (BBB) or not (classification problem). For this, we are providing different models that include Support Vector Machine (SVM), Random Forests (RF), Multi Linear Regression, among others. Predicting this properties will help the user to speed up the process for the pre-clinical trials in regards of drug discovery.
Ligand Net
Ligand Net is a tool which combines different machine learning models into one platform for the prediction of the state of the ligands either actives or decoys. If the ligands bind to the protein it is classified as active, else as decoy. This tool provides the user the feasibility to choose different parameters from data preparation to model selection. For example, for feature generation user can use Extended Circular Fingerprint, Topological Pharmacophore Atomic Triplets Fingerprint (TPATF), Topological Pharmacophore Atom Pairs Fingerprint etc. Similarly, for data splitting, user will have the option of Random splitting, Index splitting, Shuffle splitting, among others.
Ligand Net is a tool which combines different machine learning models into one platform for the prediction of the state of the ligands either actives or decoys. If the ligands bind to the protein it is classified as active, else as decoy. This tool provides the user the feasibility to choose different parameters from data preparation to model selection. For example, for feature generation user can use Extended Circular Fingerprint, Topological Pharmacophore Atomic Triplets Fingerprint (TPATF), Topological Pharmacophore Atom Pairs Fingerprint etc. Similarly, for data splitting, user will have the option of Random splitting, Index splitting, Shuffle splitting, among others.
Non-Covalent interactions
We are interested in studying role of weak non-covalent interactions in molecular recognition and develop scoring functions to incorporate such interactions.
Computer-aided drug discovery tools development
We are working in bringing theory to computation. We are developing several computational tools that can be used in computer aided drug discovery research. Please see software page of this website for our recently developed software.
GPU acceleration
Computational tasks can be time taking. We are exploiting the computational power available on today’s graphics processing units (GPUs) for accelerating these tasks considerably. During the past few years, GPUs have become an alternative to purely CPU-based high-performance computing. Our goal is to expedite the process by running the calculations parallely using the Graphical Processing Units. Our lab is currently involved int he development of GPU enabled virtual screening (molecular docking) software.
Rule Based Modeling
Rule-based modeling is a modeling approach that uses a set of rules that indirectly specifies a mathematical model. The rule-set can either be translated into a model such as Markov chains or differential equations, or be treated using tools that directly work on the rule-set in place of a translated model, as the latter is typically much bigger. Rule-based modeling is especially effective in cases where the rule-set is significantly simpler than the model it implies, meaning that the model is a repeated manifestation of a limited number of patterns. An important domain where this is often the case is biochemical models of living organisms. we are working with our collaborators at Northern Arizona University and Los Alamos National Laboratory in developing specialized software tools for modeling cellular information processing and decision making.
We are working in bringing theory to computation. We are developing several computational tools that can be used in computer aided drug discovery research. Please see software page of this website for our recently developed software.
GPU acceleration
Computational tasks can be time taking. We are exploiting the computational power available on today’s graphics processing units (GPUs) for accelerating these tasks considerably. During the past few years, GPUs have become an alternative to purely CPU-based high-performance computing. Our goal is to expedite the process by running the calculations parallely using the Graphical Processing Units. Our lab is currently involved int he development of GPU enabled virtual screening (molecular docking) software.
Rule Based Modeling
Rule-based modeling is a modeling approach that uses a set of rules that indirectly specifies a mathematical model. The rule-set can either be translated into a model such as Markov chains or differential equations, or be treated using tools that directly work on the rule-set in place of a translated model, as the latter is typically much bigger. Rule-based modeling is especially effective in cases where the rule-set is significantly simpler than the model it implies, meaning that the model is a repeated manifestation of a limited number of patterns. An important domain where this is often the case is biochemical models of living organisms. we are working with our collaborators at Northern Arizona University and Los Alamos National Laboratory in developing specialized software tools for modeling cellular information processing and decision making.