Find open-source science resources

# DrugGen 2: A disease-aware language model for enhancing drug discovery DrugGen-2 is a disease‑aware language model specialized for generating drug-like SMILES structures based on both disease pathways and protein sequence.

Plain-text, git-tracked electronic lab notebook (ELN) for reproducible bioinformatics — threads your R & Python figures into living lab notes with full provenance. Built for single-cell / CyTOF / flow cytometry; works with Obsidian, Quarto & Jupyter.

World's first fully open, accelerated weather AI software stack with Medium Range forecasting and Nowcasting models using generative AI (January 2026)

Fast, interactive, multi-dimensional image viewer for Python, foundational platform for scientific imaging AI with a rich plugin ecosystem integrating deep learning segmentation, object tracking, and microscopy analysis workflows (2.6K+ stars)

High-performance, GPU-accelerated library for key computational chemistry tasks including molecular similarity, conformer generation, and geometry relaxation, designed to accelerate drug-discovery and molecular-modeling workflows (264+ stars, Apache 2.0)

Open-source, all-atom biomolecular foundation model that turns co-folding into a scalable engine for structure prediction, design, and optimization across proteins, nucleic acids, and small molecules in drug discovery; ranked first on PXMeter-AB, FoldBench-AB, and 2026ARK-AB antibody-antigen benchmarks (263+ stars, Apache 2.0)

💻 Github | 📄 Report (Coming soon...) | 🚀 Demo

!image

Open-source AI workbench for scientific research that automates the full research loop — literature review, hypothesis generation, code writing, experiment execution, database querying, and report writing — with 290+ skills, specialized research agents, and a browser-based workspace (1453+ stars, Apache 2.0, 2026)

Turn any AI agent into an AI Scientist. The #1 Agent Skills library for science with 140+ ready-to-use skills and 100+ scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Antigravity, and the open Agent Skills standard (K-Dense-AI, 26K+ stars, 2025)

A 1.7B-parameter causal language model distilled from Qwen3-30B-A3B on 6,122 STEM chain-of-thought samples using discrepancy-informed knowledge distillation. The training objective emphasizes proof structure, detects reasoning pivot tokens through token-level divergence dynamics, smooths…

Next-generation benchmark for data-driven global weather models with standardized evaluation framework and curated datasets for ML forecasting (Google Research, 2024)

GO Rules are a way of documenting the set of filters and reports that should apply to GAF annotation data. Some rules are expressed as SPARQL on a triplestore, some are code in the GAF parsing software, ontobio.

> NEXUS domain specialist for medical Q&A and clinical reasoning — lightweight & uncensored.

Machine learning and statistical learning for neuroimaging in Python, providing easy-to-use tools for fMRI and MRI analysis including decoding, connectivity estimation, and parcellation with seamless scikit-learn integration (INRIA Parietal team, 1.4K+ stars)

> NEXUS domain specialist for medical Q&A and clinical reasoning — lightweight & uncensored.

Machine learning interatomic potentials

A library and command-line tool for building and analyzing complex homogeneous microkinetic models from quantum chemistry calculations, with support for quasi-harmonic thermochemistry, quantum tunnelling corrections, molecular symmetries and more.

MEG and EEG.

Peak Detection in Mass Spectrometry data is one of the important preprocessing steps. The performance of peak detection affects subsequent processes, including protein identification, profile alignment and biomarker identification. Using Continuous Wavelet Transform (CWT), this package provides a reliable algorithm for peak detection that does not require any type of smoothing or previous baseline correction method, providing more consistent results for different spectra. See <doi:10.1093/bioinformatics/btl355} for further details.

A quantum chemistry package written in Python.

NVIDIA's open-source platform for building and adapting biological AI models at scale, bundling ESM-2, Geneformer, MolMIM and DNA embedding models with recipes for single-GPU to multi-node training (2025)

An interface to the fast-access storage format for VCF data provided in SeqArray, with tools for common operations and analysis.

A collection of object-oriented software tools for problems involving chemical kinetics, thermodynamics, and transport processes.

Probabilistic framework for inferring cell fate decisions and trajectory dynamics from multi-view single-cell data using Markov chains and machine learning, integrating RNA velocity, pseudotime, and metabolic labeling to predict differentiation paths and terminal states (scverse/Theis Lab, 449+ stars, BSD 3-Clause)

Universal components for differentiable scientific computing, packaging heterogeneous scientific tools into self-contained, portable, gradient-propagating components with auto-generated schemas, CLI/REST API/Python SDK interfaces, and reproducible deployment across local, cloud, and HPC environments (105+ stars, Apache 2.0)

Distilling Boltz: Flow Maps for Fast All-Atom Cofolding

seqlib is a type-safe Rust library for working with DNA and RNA sequences.

Modular Python suite for Neuro-AI research across all modalities, providing efficient data loaders (NeuralSet), curated datasets (NeuralFetch), scalable training (NeuralTrain), and unified benchmarking (NeuralBench) for building and evaluating neuroscience foundation models (Meta FAIR, 270+ stars, MIT License, 2026)

OpenProteo is the open-source Rust stack for proteomics raw-file access. It reads Thermo, Bruker, and Waters acquisitions through a single API (via the sibling OpenTFRaw, OpenTimsTDF, and OpenWRaw readers), converts them to PSI-MS mzML 1.1.0 with a canonical writer, and provides a zero-copy read_arrow() API (enabled by default) that loads directly into Polars or Pandas via PyArrow. No vendor SDKs, no Windows-only DLLs, no binary blobs in the release pipeline. Includes a one-shot vendor2mzml CLI.

OpenWRaw is a standalone, cross-platform reader for Waters MassLynx .raw acquisition directories, implemented in pure Rust with no dependency on vendor DLLs. Python bindings built on PyO3 expose functions, scans, and ion-mobility data as native Python objects from Waters QTof and SYNAPT instrument families, ready to be assembled into a Pandas or Polars DataFrame.

OpenTimsTDF is a standalone, cross-platform reader for Bruker timsTOF .tdf and .tdf_bin acquisition files, implemented in pure Rust with no dependency on vendor SDKs. Python bindings built on PyO3 expose frame, scan, and peak data as native Python objects, providing ion-mobility-aware access that can be assembled into a Pandas or Polars DataFrame.

OpenTFRaw is a standalone, cross-platform reader for Thermo Fisher Scientific .raw mass-spectrometry files, implemented in pure Rust with no dependency on vendor DLLs or .NET. Python bindings built on PyO3 return NumPy arrays for spectral data, straightforward to load into Pandas or Polars. Covers format versions 8 through 66 (LCQ Classic through Orbitrap Astral and modern TSQ instruments), supporting both centroid and profile spectra.

Open-source implementation of AlphaEvolve's evolutionary coding agent paradigm, enabling LLMs to autonomously discover and optimize algorithms through iterative evolution, matching the approach behind DeepMind's breakthrough matrix multiplication discovery (6.2K+ stars, 2025)

Local-first, open-source AI workbench for scientists — an open alternative to Claude Science (by ai4s-research, maintainers of this list; TypeScript, MIT, 2026)

Modern LLM-native agent simulation platform for social science research and experimental design, providing a flexible framework for creating and managing intelligent agents in simulated environments (Tsinghua FIB Lab, 984+ stars, 2025)

Offline-first scientific writing workspace powered by Claude, integrating LaTeX, Python, and 100+ scientific skills with local execution, Zotero integration, and privacy-focused design (2026)

The Zarr specification defines a format for chunked, compressed, N-dimensional arrays. It's design allows efficient access to subsets of the stored array, and supports both local and cloud storage systems. Rarr aims to implement this specification in R with minimal reliance on an external tools or libraries.

### Model Overview TabPFN-3 is a transformer-based foundation model that uses in-context-learning to solve tabular prediction problems in a forward pass. Inference code can be found at https://github.com/PriorLabs/TabPFN. More details can be found in the Model Report.

Research ecosystem for rigorous and trustworthy AI scientists — a protocol and skill bundle that makes autonomous research verifiable, crystallized, and observable through structured, machine-executable research artifacts and five agent skills for research management, compilation, verification, visualization, and publication (ARA-Labs, 447+ stars, MIT License, 2026)

This repository contains LoRA finetunes of DiffusionGemma (image-conditioned discrete-diffusion LLM) for radiology visual question answering, each paired with an autoregressive Gemma-4 finetune as a controlled baseline. It corresponds to the paper Discrete Diffusion Language Models for Interactive…

Interactive browser for the complete Human Phenotype Ontology (~19,800 terms), with a graph-based term explorer and a clinical profile analyzer for phenotype similarity, differential diagnosis, and gene prioritization.

Open-source framework for building physics-ML models at scale (renamed from Modulus, 2025)

markeR is an R package that provides a modular and extensible framework for the systematic evaluation of gene sets as phenotypic markers using transcriptomic data. The package is designed to support both quantitative analyses and visual exploration of gene set behaviour across experimental and clinical phenotypes. It implements multiple methods, including score-based and enrichment approaches, and also allows the exploration of expression behaviour of individual genes. In addition, users can assess the similarity of their own gene sets against established collections (e.g., those from MSigDB), facilitating biological interpretation.

GPU-accelerated differentiable physics simulation engine built on NVIDIA Warp, supporting rigid/soft body, cloth, and gradient-based optimization for scientific ML, initiated by Disney Research, DeepMind, and NVIDIA (Linux Foundation, Apache 2.0, 2025)

Composite-objective protein design framework integrating Boltz, AlphaFold2, OpenFold3, ProteinMPNN, and ESM via JAX-based gradient optimization over continuous relaxed sequence space for multi-property binder design (319+ stars, MIT License, 2025)

LLM papers for scientific discovery

Toolkit for large-scale whole-slide image processing supporting 22+ patch encoders (UNI, CONCH, Virchow, H-Optimus-0, etc.), slide encoders (TITAN, GigaPath, PRISM, CHIEF, Madeleine, Feather), tissue segmentation, and multi-GPU inference with end-to-end pipeline and smart resume for standardized deployment of computational pathology foundation models (Mahmood Lab, Harvard Medical School, 553+ stars)