/tools

This repository provides scripts and data for predicting protein stability using deep learning representations. It includes a model for rapid predictions and tools for data visualization and analysis, making it useful for researchers in molecular biology and computational chemistry.

This repository provides data and notebooks for fine-tuning protein language models to enhance predictions across diverse tasks. It includes training datasets and examples for generating embeddings and training models, making it a useful resource for molecular machine learning in protein-related applications.

Wazy is a tool for Bayesian optimization of amino acid sequences, allowing users to design peptides that bind to specific proteins. It utilizes pretrained models to predict the properties of sequences and optimize their design through an interactive interface.

The 'walk-jump' repository provides an open-source implementation of discrete Walk-Jump Sampling (dWJS) for protein design. It includes functionalities for training models and sampling, aimed at discovering and optimizing protein sequences.

ProCyon is an open-source multimodal foundation model designed to predict protein phenotypes across various scales. It includes capabilities for drug-binding domain prediction and provides benchmarking models for systematic evaluation against other methods.

CaLM is a codon adaptation language model designed to provide embeddings for DNA sequences, specifically aimed at enhancing protein engineering efforts. It allows users to embed sequences, which can be useful for predicting and optimizing protein properties.

SeqDance and ESMDance are biophysics-informed protein language models that predict dynamic properties of proteins based on their sequences. They leverage molecular dynamics data to inform mutation effects on protein folding stability, making them valuable tools in protein design and property prediction.

RamaNet is a tool that utilizes machine learning and PyRosetta to autonomously generate novel helical protein structures. It performs de novo design by creating a topology and optimizing the sequence to fit, while also providing datasets for training the neural network.

PPIformer is a machine learning model that predicts the effects of mutations on protein-protein interactions by estimating binding energy changes. It utilizes pre-training and fine-tuning on specific datasets to enhance its predictive capabilities for protein interactions.

Origin-1 is a generative AI platform designed for the de novo design of antibodies targeting novel epitopes. It includes data on binding affinity and optimization results, making it a valuable resource for antibody development in molecular biology.

CARBonAra is a deep learning framework that facilitates the design of protein sequences by utilizing atomic coordinates and context-aware generation methods. It allows for the integration of molecular environments, including non-protein entities, enhancing the control over protein engineering and design.

This repository implements a machine learning-guided framework for protein design through directed evolution. It utilizes large language models to predict fitness scores and generate novel protein sequences, streamlining the optimization process in protein engineering.

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.

This repository provides code for fine-tuning protein language models using a parameter-efficient approach. It includes configuration files and scripts for running experiments related to protein interactions and symmetry, making it a useful tool for researchers in molecular biology.

This repository provides a course on protein language modeling, covering topics such as sequence analysis, model training, and protein design. It includes hands-on notebooks for predicting features and classifying protein sequences using embeddings generated by language models.

LLM4Mol is a repository that explores the application of large language models in molecular design and protein research. It serves as a hub for studies and techniques that leverage AI to advance understanding in molecular properties and material science.

MapDiff is a PyTorch implementation of a deep diffusion model designed to improve the inverse protein folding task. It predicts feasible amino acid sequences from 3D protein backbone structures, contributing to the field of protein design.

The BA-DDG repository provides an implementation of a Boltzmann-Aligned Inverse Folding Model that predicts the effects of mutations on protein-protein interactions. It utilizes datasets like SKEMPI v2 and offers both supervised and unsupervised inference methods.

DeepNano is a tool designed for predicting interactions between nanobodies and antigens using ensemble deep learning and protein language models. It provides code and model weights for various prediction tasks related to protein-protein interactions.

GeoAB is a tool designed for realistic antibody design and reliable affinity maturation. It provides a framework for training models that can generate and optimize antibody structures, utilizing datasets for evaluation and refinement.

SiamDiff is a codebase for a diffusion-based pre-training algorithm that enhances protein structure encoders. It improves performance on tasks such as protein-protein interaction prediction and function annotation by learning effective representations from protein sequences and structures.

π-PrimeNovo is a deep learning model designed for accurate and efficient de novo peptide sequencing. It addresses challenges in peptide identification from mass spectrometry data, making it a valuable tool in molecular biology and protein design.

proseLM is a protein language model that adapts to structure-conditioned sequence generation, facilitating the design of proteins based on their structural information. This tool aims to enhance the capabilities in protein design and engineering through advanced machine learning techniques.

ProteinReDiff is a framework that utilizes equivariant diffusion-based generative models to redesign ligand-binding proteins. It allows for the generation of high-affinity proteins based on initial sequences and ligand SMILES, facilitating advancements in drug discovery and protein engineering.