/tools

The S3F repository provides a multimodal representation learning framework for predicting protein fitness. It integrates various protein features and is evaluated using the ProteinGym benchmark, making it a valuable tool for protein design and property prediction.

Runs N' Poses is a benchmark dataset designed for evaluating protein-ligand co-folding prediction methods. It includes various metrics and data formats to facilitate machine learning applications in molecular biology, particularly for assessing the generalization of prediction models.

This repository provides code and datasets for analyzing IC50 and Ki values from various sources, highlighting the noise in these measurements. It includes curated datasets from ChEMBL32 and tools for generating results related to molecular property prediction.

SPECTRA is a Python toolkit that provides a spectral framework for evaluating the generalizability of biomedical AI models, particularly in the context of molecular datasets. It allows users to define spectral properties and generate train-test splits to assess model performance across varying degrees of data overlap.

GenScore is a generalized scoring framework for protein-ligand interactions that extends RTMScore. It provides capabilities for scoring, ranking, and screening on multiple datasets, making it useful for molecular property prediction and docking tasks.

The COMP6 repository contains a benchmark suite for assessing the performance of machine-learning molecular potentials. It includes results and methodologies for evaluating the accuracy of these models in predicting molecular properties.

The Protein-Language-Models repository provides a systematic review of protein language models, covering their architectures, evaluation metrics, and relevant datasets. It also introduces tools for ongoing research in the field of protein modeling and analysis.

MolRL-MGPT is a code repository for a NeurIPS 2023 paper that presents a method for de novo drug design using multiple GPT agents in a reinforcement learning framework. It incorporates large molecular datasets and benchmarks for evaluating the generated molecules, making it a relevant tool in the field of molecular design and drug discovery.

TFTraj is a tool for analyzing molecular dynamics trajectories using TensorFlow, specifically implementing routines for calculating root-mean-square deviation (RMSD). It includes benchmarks for performance comparison with other methods, making it useful for researchers in molecular simulations.

AutoPeptideML is an AutoML system designed to help researchers build trustworthy models for predicting peptide bioactivity. It provides tools for model building, prediction, and benchmarking, making it accessible for users without prior machine learning expertise.

ChemIQ is a benchmark for evaluating the chemical intelligence of large language models by testing their ability to interpret molecular structures and perform chemical reasoning tasks. It includes a variety of questions related to molecular properties and transformations, making it relevant to the field of computational chemistry.

FAHBench is the official Folding@Home benchmark that utilizes the OpenMM molecular dynamics engine to evaluate performance on various OpenCL and CUDA-capable devices. It is designed to facilitate research in protein dynamics by providing a standardized benchmarking framework.

USPNet is a tool designed to predict signal peptides in protein sequences using a deep protein language model. It provides a benchmark set for evaluation and allows users to process and predict using their own protein data.

PDBench is a dataset and software package that evaluates fixed-backbone sequence design algorithms for proteins. It includes a benchmark set of protein structures and provides metrics for assessing the performance of various design models.

GSCDB is a comprehensive benchmark database containing 137 datasets with detailed molecular properties such as reaction energies and barrier heights. It serves as a platform for validating density functional approximations and supports the development of machine-learned functionals in computational chemistry.

Chemprop benchmarking scripts and data provide a framework for evaluating the performance of Chemprop, a message passing neural network designed for predicting molecular properties. The repository includes various benchmarks and datasets that facilitate the assessment of molecular property prediction models.

LeMat-GenBench is a comprehensive benchmarking framework designed to evaluate crystal generative models across various metrics such as validity, diversity, and stability. It facilitates the assessment of material generation models, making it relevant for molecular design and optimization in computational chemistry.

PREFER is a benchmarking and property prediction framework that automates the evaluation of different molecular representations and machine learning models for predicting molecular properties. It supports various models and configurations, allowing users to predict properties like solubility and logD using data-driven molecular representations.

This tool implements the Confidence Bootstrapping procedure for enhancing protein-ligand docking predictions. It includes pretrained models and datasets for benchmarking, making it useful for researchers in molecular docking and drug discovery.

The 'conformer-benchmark' repository provides data and scripts for assessing conformer energies using electronic structure and machine learning methods. It serves as a living benchmark to include new methods and data for evaluating conformer relative energies in molecular systems.

RetroCast is a comprehensive toolkit designed for standardizing, scoring, and analyzing multistep retrosynthesis models. It provides a unified framework for evaluating different retrosynthesis algorithms, facilitating rigorous comparisons and improving reproducibility in molecular design.

AEV-PLIG is a tool that utilizes a graph neural network to predict the binding affinity of protein-ligand complexes based on their 3D structures. It benchmarks its performance against established datasets and demonstrates how to train and use the model for predictions.

PIGNet2 is a deep learning-based model that predicts protein-ligand interactions and binding affinities, facilitating virtual screening in drug discovery. It includes training and benchmarking scripts, making it a comprehensive tool for evaluating molecular interactions.

The 'protein-uq' repository provides tools for benchmarking uncertainty quantification methods in protein engineering. It includes functionalities for training models that predict protein fitness and function, utilizing active learning and Bayesian optimization techniques.