Find open-source science resources

Neural Network Force Field based on PyTorch.

multiHiCcompare provides functions for joint normalization and difference detection in multiple Hi-C datasets. This extension of the original HiCcompare package now allows for Hi-C experiments with more than 2 groups and multiple samples per group. multiHiCcompare operates on processed Hi-C data in the form of sparse upper triangular matrices. It accepts four column (chromosome, region1, region2, IF) tab-separated text files storing chromatin interaction matrices. multiHiCcompare provides cyclic loess and fast loess (fastlo) methods adapted to jointly normalizing Hi-C data. Additionally, it provides a general linear model (GLM) framework adapting the edgeR package to detect differences in Hi-C data in a distance dependent manner.

Universal Cell Embeddings: zero-shot single-cell foundation model pretrained on 36M cells across 11M species, learning cross-species gene function representations via a protein-language-model-informed token space; enables zero-shot cell type annotation, embedding, and integration of unseen datasets and species without fine-tuning (338+ stars, MIT License)

DeepTaxa is a hybrid CNN-BERT deep learning framework for multi-rank taxonomic classification of 16S rRNA gene sequences. It predicts all seven Linnaean ranks from domain to species in a single forward pass and provides pre-trained checkpoints for full-length 16S and V3-V4 amplicons.

standR is an user-friendly R package providing functions to assist conducting good-practice analysis of Nanostring's GeoMX DSP data. All functions in the package are built based on the SpatialExperiment object, allowing integration into various spatial transcriptomics-related packages from Bioconductor. standR allows data inspection, quality control, normalization, batch correction and evaluation with informative visualizations.

A package for creating fast and accurate interatomic potentials.

Long-context generative genomic foundation model using 6-mer tokenization for DNA sequence modeling and generation, with v2 model families for prokaryote and eukaryote genomes and pretrained weights available on HuggingFace (GenerTeam, 460+ stars, MIT License, 2025-2026)

LLM papers for scientific discovery

"rhinotypeR" is designed to automate the comparison of sequence data against prototype strains, streamlining the genotype assignment process. By implementing predefined pairwise distance thresholds, this package makes genotype assignment accessible to researchers and public health professionals. This tool enhances our epidemiological toolkit by enabling more efficient surveillance and analysis of rhinoviruses (RVs) and other viral pathogens with complex genomic landscapes. Additionally, "rhinotypeR" supports comprehensive visualization and analysis of single nucleotide polymorphisms (SNPs) and amino acid substitutions, facilitating in-depth genetic and evolutionary studies.

Family of large language models for materials research via continued pretraining of LLaMA-2/3 on ~30B materials science tokens, outperforming commercial LLMs on materials science tasks while identifying "adaptation rigidity" in overtrained models; includes MatNLP benchmark and CIF crystal generation capabilities (IIT Delhi M3RG, MIT License)

This package addresses two broad areas. It allows for in-depth analysis of spatial transcriptomic data by identifying tissue neighbourhoods. These are contiguous regions of tissue surrounding individual cells. 'CatsCradle' allows for the categorisation of neighbourhoods by the cell types contained in them and the genes expressed in them. In particular, it produces Seurat objects whose individual elements are neighbourhoods rather than cells. In addition, it enables the categorisation and annotation of genes by producing Seurat objects whose elements are genes.

Medical time series foundation model pretrained on 454B time points from heterogeneous clinical corpora spanning ICU physiological signals and hospital EHR, with continuous-time rotary positional encoding, frequency-specialized Mixture-of-Experts, and neural ODE extrapolation for zero-shot forecasting across irregular and multimodal temporal health data (Microsoft, 399+ stars, MIT License)

Gene regulatory networks model the underlying gene regulation hierarchies that drive gene expression and observed phenotypes. Epiregulon infers TF activity in single cells by constructing a gene regulatory network (regulons). This is achieved through integration of scATAC-seq and scRNA-seq data and incorporation of public bulk TF ChIP-seq data. Links between regulatory elements and their target genes are established by computing correlations between chromatin accessibility and gene expressions.

scToppR provides an easy-to-use API wrapper for the ToppGene web platform, used for gene ontology and functional enrichment research. The package also integrates visualization tools, making it a convenient tool directly connecting ToppGene to code-based workflows in R. The tool can also easily save results into different formats.

The package implements an algorithm for fast gene set enrichment analysis. Using the fast algorithm allows to make more permutations and get more fine grained p-values, which allows to use accurate stantard approaches to multiple hypothesis correction.

Cython + HTSlib == fast VCF parsing; even faster parsing than pyVCF.

Provides hurdle negative binomial models for differential expression analysis with long-read RNA-Seq data.

Provides a streamlined workflow for clustering and visualizing gene expression patterns, particularly from time-series RNA-Seq and single-cell experiments. The package is designed to integrate seamlessly within the Bioconductor ecosystem by operating directly on standard data classes such as `SummarizedExperiment` and `SingleCellExperiment`. It implements common clustering algorithms (e.g., k-means, fuzzy c-means) and generates a suite of publication-ready visualizations to explore co-expressed gene modules. Functions are also included to facilitate the visualization of clustering results derived from other popular tools.

Autonomous AI scientist research

A powerful, high-performance bioinformatics framework for discovering, evaluating, and verifying microRNA (miRNA) interactions across DNA, RNA, and protein target sequences. The mirnaprotpred package provides two core modules: SeqFinder: A discovery engine to find all potential miRNA interactions across a genome or target sequence. Validator: A targeted verification engine to test specific, user-provided miRNAs against a target sequence. Both modules are powered by a shared, rigorous biological engine that evaluates exact seed matching, wobble pairing, AU-rich context, and RNAduplex thermodynamic stability.

PhantasusLite – a lightweight package with helper functions of general interest extracted from phantasus package. In parituclar it simplifies working with public RNA-seq datasets from GEO by providing access to the remote HSDS repository with the precomputed gene counts from ARCHS4 and DEE2 projects.

This package implements a metabolic network analysis pipeline to identify an active metabolic module based on high throughput data. The pipeline takes as input transcriptional and/or metabolic data and finds a metabolic subnetwork (module) most regulated between the two conditions of interest. The package further provides functions for module post-processing, annotation and visualization.

Cross-domain foundation model for continuum dynamics trained on 19 physical scenarios spanning 63 variables, featuring adaptive compute via stride modulation and patch jittering for long-run stability (Polymathic AI, 293+ stars, MIT License)

Provides a reproducible and modular workflow for absolute microbial quantification using spike-in controls. Supports both single spike-in taxa and synthetic microbial communities with user-defined spike-in volumes and genome copy numbers. Compatible with 'phyloseq' and 'TreeSummarizedExperiment' (TSE) data structures. The package implements methods for spike-in validation, preprocessing, scaling factor estimation, absolute abundance conversion, bias correction, and normalization. Facilitates downstream statistical analyses with 'DESeq2', 'edgeR', and other Bioconductor-compatible methods. Visualization tools are provided via 'ggplot2', 'ggtree', and related packages. Includes detailed vignettes, case studies, and function-level documentation to guide users through experimental design, quantification, and interpretation.

Autonomous multi-agent AI scientist that mirrors a PhD workflow: literature review → grounded hypothesis → real ML experiments → self-critique → write-up; features a deterministic harness with frozen-metric scoring, edit allowlists, and a verified registry to make reward-hacking and hallucination impossible, plus 108 unit tests runnable without API keys or GPUs (461+ stars, MIT License, 2026)

DenoIST identifies and removes contamination in Image-based Spatial Transcriptomics data, using a transposed poisson mixture model with local neighbourhood offsets to infer genes that are likely to be due to neighbourhood contamination rather than endogenous expression.

MeLSI (Metric Learning for Statistical Inference) is a novel machine learning method for microbiome data analysis that learns optimal distance metrics to improve statistical power in detecting group differences. Unlike traditional distance metrics (Bray-Curtis, Euclidean, Jaccard), MeLSI adapts to the specific characteristics of your dataset to maximize separation between groups. The method uses an ensemble of weak learners to identify which microbial features drive group differences, providing both improved statistical power and biological interpretability through feature importance weights.

Have you ever index sorted cells in a 96 or 384-well plate and then sequenced using Sanger sequencing? If so, you probably had some struggles to either check the electropherogram of each cell sequenced manually, or when you tried to identify which cell was sorted where after sequencing the plate. Scifer was developed to solve this issue by performing basic quality control of Sanger sequences and merging flow cytometry data from probed single-cell sorted B cells with sequencing data. scifer can export summary tables, 'fasta' files, electropherograms for visual inspection, and generate reports.

Diffusion-based document OCR framework replacing autoregressive decoding with block-level parallel diffusion decoding, enabling high-accuracy text recognition in scientific PDFs (613+ stars, MIT License)

pairedGSEA makes it simple to run a paired Differential Gene Expression (DGE) and Differencital Gene Splicing (DGS) analysis. The package allows you to store intermediate results for further investiation, if desired. pairedGSEA comes with a wrapper function for running an Over-Representation Analysis (ORA) and functionalities for plotting the results.

Annotate a VCF with other VCFs/BEDs/tabixed files.

Cross-platform system optimizations for accelerating AlphaFold3 training with 1.73x speedup and 1.23x memory reduction

Learning the language of protein-protein interactions

Genomic foundation model for metagenomic and genome annotation, featuring an 8k base-pair context and 500M parameters trained on 386B base pairs of eukaryotic DNA; provides expert models and a unified CLI for prokaryotic/eukaryotic coding-sequence annotation with strong performance on Genomic Benchmarks, Nucleotide Transformer tasks, and custom Gener tasks (GenerTeam, 314+ stars, MIT License)

Automatic atomic model building program for cryo-EM maps using deep learning, enabling rapid de novo protein structure determination from electron density with high accuracy (3DEM/EMBL, 169+ stars)

Gene regulatory networks model the underlying gene regulation hierarchies that drive gene expression and observed phenotypes. Epiregulon infers TF activity in single cells by constructing a gene regulatory network (regulons). This is achieved through integration of scATAC-seq and scRNA-seq data and incorporation of public bulk TF ChIP-seq data. Links between regulatory elements and their target genes are established by computing correlations between chromatin accessibility and gene expressions.

General-purpose deep learning backbone for molecular modeling

The ReactomeGSA packages uses Reactome's online analysis service to perform a multi-omics gene set analysis. The main advantage of this package is, that the retrieved results can be visualized using REACTOME's powerful webapplication. Since Reactome's analysis service also uses R to perfrom the actual gene set analysis you will get similar results when using the same packages (such as limma and edgeR) locally. Therefore, if you only require a gene set analysis, different packages are more suited.

Implementation of the Ibex algorithm for single-cell embedding based on BCR sequences. The package includes a standalone function to encode BCR sequence information by amino acid properties or sequence order using tensorflow-based autoencoder. In addition, the package interacts with SingleCellExperiment or Seurat data objects.

Data analysis for Single File Injections(SFIs) mode LC-MS analysis. In SFIs mode, pooled samples are initially injected to serve as reference peaks for subsequent analyses. Repeated injections of individual samples are then performed at fixed time intervals using isocratic elution. This package provides the functions to analyze data from SFIs mode including peak picking and peak reassignment.

A Swiss Army knife for genome arithmetic.

A static web application presents an interactive knowledge graph of single-cell long-read RNA sequencing literature synthesized from seven source papers. Users navigate mind-tree, network graph, guided learning-path, and Sankey views linking platforms, protocols, methods, and software. A benchmark tab provides 34 question-answer pairs with category and difficulty filters, exportable as JSON or CSV for LLM and agent evaluation.

Bias factorized, base-resolution deep learning models of chromatin accessibility (chromBPNet).

Multi-agent system with Parser-Planner-Painter architecture converting `paper.pdf` to editable `poster.pptx`, outperforms GPT-4o with 87% fewer tokens

Self-evolving AI research colleague built on OpenClaw with 285+ runtime-adaptive skills across 28+ disciplines, persistent cross-session research memory, and zero-hallucination citation protocols; agent autonomously writes new SKILL.md files based on research patterns without redeployment (828+ stars, MIT License, 2026)

Transformer foundation model for tandem mass spectrometry (MS/MS) self-supervised on millions of unannotated spectra from the GeMS dataset via masked peak prediction and chromatographic retention-order objectives, producing 1024-dimensional molecular representations; achieves SOTA on spectral similarity, chemical property, and molecular fingerprint prediction, and powers the DreaMS Atlas annotating 201M+ MS/MS spectra for metabolomics and natural product discovery (Pluskal Lab, IOCB Prague & MIT, 211+ stars, MIT License)

LLM-native molecular language that represents molecules as explicit graph-based code, enabling LLMs to operate and reason on chemistry directly with 5× lower token cost and ~76-80% accuracy on novel molecules vs ~20% for SMILES; supports small molecules, polymers, and Markush structures with lossless RDKit interconversion and Claude Code/Codex agent skills (AtomFlow, arXiv:2605.16480, 281+ stars, MIT License, 2026)

A graph-based molecule modeling and chemoinformatics analysis toolkit fully implemented in Julia

Programmatic framework for designing state-switching proteins via backpropagation through compositional design constraints parameterized by structure prediction models; enables de novo design of allosteric regulators and fluorescent biosensors for arbitrary small-molecule analytes (79+ stars, MIT License, ICML 2026)

Efficient differentiable n-dimensional PDE solvers built on JAX and Equinox, shipping 46+ built-in equations with Fourier spectral methods, exponential time differencing, and full auto-differentiation for physics-based deep learning workflows (MIT, 200+ stars, 2024)