/tools

This repository develops machine learning methods to predict adverse drug reactions (ADRs) using databases like SIDER and OFFSIDES. It provides tools and models that can be utilized in the drug discovery process to enhance safety and efficacy assessments.

MultiStepRetrosynthesisTTL is a tool designed for predicting multistep retrosynthesis routes by utilizing a disconnection aware triple transformer loop approach. It integrates various models to enhance the prediction of synthetic pathways, making it a valuable resource for molecular design and synthesis in computational chemistry.

do_x3dna is a tool designed for analyzing the dynamics of DNA and RNA structures during molecular dynamics simulations. It provides functionalities to calculate structural descriptors and analyze fluctuations in molecular trajectories.

RAG-ESM is a framework that enhances pretrained protein language models by conditioning them on homologous sequences. It allows for the generation of novel protein sequences and improves predictive performance through a retrieval-augmented approach.

LEADD is a tool that employs a Lamarckian evolutionary algorithm for the design and optimization of molecules in drug discovery. It utilizes a population-based approach to evolve molecular structures by combining fragments and optimizing them based on user-defined scoring functions.

ProtProfileMD is a tool that leverages protein dynamics to enhance homology detection by predicting per-residue distributions over structural representations. It utilizes a sequence-to-3D-probability model to capture the flexibility of proteins, improving the sensitivity of remote homology detection.

PSALM is a tool designed for protein sequence annotation using advanced language models. It allows users to scan protein sequences for functional domains and provides a framework for training and evaluating models on protein data.

This repository provides scripts for training neural network models to predict binding affinities of ligands to receptors. It includes functionalities for clustering data for cross-validation and generating unique ligand poses, making it a useful tool for molecular property prediction in drug discovery.

The MongoDB Chemistry repository provides an implementation for chemical similarity searches using MongoDB, along with performance analysis. It allows users to load molecular data and perform similarity searches, which is useful in various applications within computational chemistry.

Uni-Mol Tools is an easy-to-use auto-ML tool designed for predicting molecular properties and representations. It allows users to train models for classification and regression tasks on molecular data, utilizing SMILES and other formats for input.

LaM-SLidE is a tool for latent space modeling of spatial dynamical systems, particularly in the context of molecular dynamics simulations. It provides a framework for training models that can handle various molecular types, including peptides, and facilitates the analysis of molecular properties through simulation.

DiffDock is a Colab implementation of a state-of-the-art molecular docking method that allows users to perform docking simulations for single and multiple protein-ligand complexes. It provides example data files and notebooks to facilitate the docking process, making it a useful tool for researchers in molecular biology and computational chemistry.

Neuraldecipher is a tool that implements a method for reverse-engineering extended-connectivity fingerprints (ECFPs) back to their corresponding molecular structures. It utilizes deep learning techniques to facilitate the generation and analysis of molecular representations, making it useful for cheminformatics applications.

IDSL_MINT is a deep learning framework designed to interpret raw mass spectrometry data, enabling the prediction of molecular fingerprints and structures from MS/MS spectra. It utilizes transformer models to translate mass spectra into molecular descriptors and canonical SMILES, making it a valuable tool for cheminformatics and molecular property prediction.

OM-TPS implements a method for efficient transition path sampling by repurposing pre-trained generative models. It utilizes the Onsager-Machlup functional to find maximum likelihood paths in molecular systems, making it applicable for studying molecular dynamics.

The Chemical Graph Series is an educational resource that teaches users how to represent molecules as graphs and apply deep learning techniques, specifically graph neural networks, to predict chemical properties. It covers foundational concepts in cheminformatics and progresses to advanced modeling techniques, including property prediction using real datasets.

The antiberty-pytorch repository provides an unofficial re-implementation of the AntiBERTy model, which is designed to analyze and predict properties of antibody sequences using a language model approach. It includes a dataset preparation pipeline for working with observed antibody sequences, facilitating research in antibody affinity maturation.

Geo-DEG is a tool designed for data-efficient molecular property prediction by utilizing a learnable hierarchical grammar that generates molecules. It employs graph neural networks to predict properties based on the induced geometry of molecular graphs, outperforming various baseline models.

PyBILT is a Python toolkit developed for the analysis of lipid bilayer molecular simulation trajectories. It includes various analyses such as mean squared displacement, diffusion coefficients, and bilayer thickness, making it a valuable tool for researchers in molecular dynamics.

DeepSeqPan is a deep convolutional neural network model designed for predicting the binding affinity of peptides to MHC I molecules. It includes datasets for training and evaluation, as well as trained models for use in further research.

DeepGS is a software tool designed for predicting drug-target binding affinity using deep representation learning techniques. It processes molecular and protein sequence data to generate predictions, making it a valuable resource in the field of drug discovery.

GPUGA is a tool designed for empirical potential fitting using a genetic algorithm, optimized for GPU performance. It allows for efficient modeling of molecular interactions, particularly in the context of materials like diamond silicon.