Find open-source science resources

Google DeepMind's diffusion-based ensemble weather forecasting model at 0.25° resolution, outperforming ECMWF ENS on 97.2% of targets up to 15 days ahead, with open-source code and weights (Nature 2024)

The MsQuality provides functionality to calculate quality metrics for mass spectrometry-derived, spectral data at the per-sample level. MsQuality relies on the mzQC framework of quality metrics defined by the Human Proteom Organization-Proteomics Standards Initiative (HUPO-PSI). These metrics quantify the quality of spectral raw files using a controlled vocabulary. The package is especially addressed towards users that acquire mass spectrometry data on a large scale (e.g. data sets from clinical settings consisting of several thousands of samples). The MsQuality package allows to calculate low-level quality metrics that require minimum information on mass spectrometry data: retention time, m/z values, and associated intensities. MsQuality relies on the Spectra package, or alternatively the MsExperiment package, and its infrastructure to store spectral data. Additionally, MsQuality supports Chromatograms objects from the Chromatograms package for chromatographic quality metrics.

EPOS-DCAT-AP is an extension of the DCAT Application Profile for Research Infrastructures in Environmental and the solid-Earth domains.

Google Colab-based no-code toolbox democratizing deep learning in microscopy for biologists without programming experience, enabling AI-powered image segmentation, denoising, super-resolution, and object tracking across diverse imaging modalities (Henriques Lab, 640+ stars)

This repository contains DNA-sequence modeling resources associated with the basal ganglia (BG) cell atlas package. It serves as a centralized entry point for sequence-based regulatory analyses across multiple companion studies.

A differential abundance analysis for the comparison of two or more conditions. Useful for analyzing data from standard RNA-seq or meta-RNA-seq assays as well as selected and unselected values from in-vitro sequence selections. Uses a Dirichlet-multinomial model to infer abundance from counts, optimized for three or more experimental replicates. The method infers biological and sampling variation to calculate the expected false discovery rate, given the variation, based on a Wilcoxon Rank Sum test and Welch's t-test (via aldex.ttest), a Kruskal-Wallis test (via aldex.kw), a generalized linear model (via aldex.glm), or a correlation test (via aldex.corr). All tests report predicted p-values and posterior Benjamini-Hochberg corrected p-values. ALDEx2 also calculates expected standardized effect sizes for paired or unpaired study designs. ALDEx2 can now be used to estimate the effect of scale on the results and report on the scale-dependent robustness of results.

Allen Institute for AI's global geospatial foundation model for satellite imagery analysis, enabling large-scale mapping of buildings, wind turbines, trees, and land cover from Sentinel-2 data with open-source weights and inference tools (2024)

🧬 BioReason-ProAdvancing Protein Function Prediction withMultimodal Biological Reasoning

🧬 BioReason-ProAdvancing Protein Function Prediction withMultimodal Biological Reasoning

The package implements a method for normalising microarray intensities from single- and multiple-color arrays. It can also be used for data from other technologies, as long as they have similar format. The method uses a robust variant of the maximum-likelihood estimator for an additive-multiplicative error model and affine calibration. The model incorporates data calibration step (a.k.a. normalization), a model for the dependence of the variance on the mean intensity and a variance stabilizing data transformation. Differences between transformed intensities are analogous to "normalized log-ratios". However, in contrast to the latter, their variance is independent of the mean, and they are usually more sensitive and specific in detecting differential transcription.

First agentic framework for weather science, pairing an LLM with ZephyrusWorld (a code-execution environment exposing WeatherBench 2 data, geolocation, forecasting, simulation, and climatology tools) and ZephyrusBench (2,230 Q&A pairs across 49 weather-science tasks); outperforms text-only baselines by up to 44.2 percentage points (UC San Diego Rose-STL-Lab, 99+ stars, MIT License, 2026)

Provides R bindings for HiSpa, a hierarchical Bayesian model for inferring three-dimensional chromatin structures from Hi-C contact matrices using Markov Chain Monte Carlo (MCMC) sampling. The package implements a cluster-based hierarchical approach that efficiently handles large-scale Hi-C datasets. It uses Rcpp and RcppArmadillo for efficient C++ integration with the original HiSpa C++ implementation, enabling fast computation of chromatin structure inference through parallel MCMC sampling.

Unified ML/DL framework for drug discovery workflows, integrating RDKit, DeepChem, and scikit-learn with SHAP explainability

Provides an interface to build a unified database of genomic annotations and their coordinates (gene, transcript and exon levels). It is aimed to be used when simple tab-delimited annotations (or simple GRanges objects) are required instead of the more complex annotation Bioconductor packages. Also useful when combinatorial annotation elements are reuired, such as RefSeq coordinates with Ensembl biotypes. Finally, it can download, construct and handle annotations with versioned genes and transcripts (where available, e.g. RefSeq and latest Ensembl). This is particularly useful in precision medicine applications where the latter must be reported.

This package provides functions used in Seqtometry (Kousnetsov et al. 2024), a method for analyzing single cell (scRNA-seq or scATAC-seq) data via signature (gene set) enrichment scores. The Seqtometry scores may be useful for annotating or characterizing cells, either in a flow cytometry like workflow (where scores are standalone features used for progressive partitoning as described in the Seqtometry publication) or in a cluster-based workflow (as features of clusters). The exported impute function (a port of Python's MAGIC-impute, van Dijk et al. 2018), may also be useful for single cell analysis on its own.

Andrej Karpathy's autonomous LLM research framework: AI agent runs overnight experiments on a real training setup, auto-editing code→5min training→evaluation in a loop, ~100 experiments per night on a single GPU

Toolkit for linearizing academic PDFs into LLM-ready text with high accuracy and structure preservation, optimized for scientific literature extraction

RNAshapeQC provides coverage-shape-based quality control (QC) metrics for mRNA-seq and total RNA-seq data. It supports per-gene pileup construction from BAM files as well as toy datasets for quick-start examples. The package implements protocol-specific metrics, including decay rate (DR), degradation score (DS), mean coverage depth (MCD), window coefficient of variation (wCV), area under the curve (AUC), and shape-based sample-level indices. RNAshapeQC also includes HPC-friendly functions for per-gene batch processing and cross-study pileup generation. This package enables interpretable, protocol-specific QC assessments for diverse RNA-seq workflows.

CPSM provides a comprehensive computational pipeline for predicting survival probability and risk groups in cancer patients. The package includes steps for data preprocessing, training/test split, and normalization. It enables feature selection using univariate survival analysis and computes a LASSO-based prognostic index (PI) score. CPSM supports the development of predictive models using various feature sets and offers a suite of visualization tools, including survival curves based on predicted probabilities, barplots for predicted mean and median survival times, KM plots overlaid with individual survival predictions, and nomograms for estimating 1-, 3-, 5-, and 10-year survival probabilities. This makes CPSM a versatile tool for survival analysis in cancer research.

End-to-end semi-automated scientific discovery system that designs, iterates, and analyzes code-based experiments via LLM-as-a-mutator over scientific articles and code examples; auto-creates, runs, and debugs experiment code in containers and writes meta-analysis reports (339+ stars, Apache 2.0)

Quantification and differential analysis of mass-spectrometry proteomics data, with probabilistic recovery of information from missing values. Avoids the need for imputation. Estimates the detection probability curve (DPC), which relates the probability of successful detection to the underlying log-intensity of each precursor ion, and uses it to incorporate missing values into protein quantification and into subsequent differential expression analyses. The package produces objects suitable for downstream analysis in limma. The package accepts precursor (or peptide) intensities including missing values and produces complete protein quantifications without the need for imputation. The uncertainty of the protein quantifications is propagated through to the limma analyses using variance modeling and precision weights, ensuring accurate error rate control. The analysis pipeline can alternatively work with PTM or protein level data. The package name "limpa" is an acronym for "Linear Models for Proteomics Data".

Provides functions for counting reads from high-throughput sequencing screen data (e.g., CRISPR, shRNA) to quantify barcode abundance. Currently supports single barcodes in single- or paired-end data, and combinatorial barcodes in paired-end data.

Large-scale benchmark suite for protein fitness prediction and design, aggregating 200+ deep mutational scanning assays and clinical variant datasets across diverse protein families and taxa, with standardized zero-shot and supervised leaderboards for variant effect prediction, mutation effect prediction, and protein language model evaluation (OATML & Marks Lab, NeurIPS 2023 Spotlight, Datasets & Benchmarks)

Automated code generation from machine learning research papers into runnable implementations (4.5K+ stars, 2025)

Low-cost, modular self-driving laboratory platform democratizing autonomous chemical experimentation with open control software, device CAD/PCB files, and example optimization campaigns (Noël Research Group, University of Amsterdam, Apache 2.0, 2026)

### Model Overview TabPFN-2.5 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.

recoup calculates and plots signal profiles created from short sequence reads derived from Next Generation Sequencing technologies. The profiles provided are either sumarized curve profiles or heatmap profiles. Currently, recoup supports genomic profile plots for reads derived from ChIP-Seq and RNA-Seq experiments. The package uses ggplot2 and ComplexHeatmap graphics facilities for curve and heatmap coverage profiles respectively.

MarkushGrapher-2 is an end-to-end multimodal model for recognizing chemical structures from patent document images. It jointly encodes vision, text, and layout information to convert Markush structure images into machine-readable CXSMILES representations.

AlphaFold fine-tuned with flow matching for generating protein conformational ensembles, covering both experimental PDB states and molecular dynamics ensembles at physiological temperatures; includes ESMFlow variant (MIT, 526+ stars, 2024)

This package implements algorithms and data structures for performing gene expression signature (GES) searches, and subsequently interpreting the results functionally with specialized enrichment methods.

# or·a·cle /ˈôrəkəl/ — a source of wise counsel; one who provides authoritative knowledge. From Latin ōrāculum, meaning divine announcement. In computer science, an oracle is a black box that always returns the correct answer — you don't ask it how it knows, you ask and it answers.

EVA is a generative foundation model for universal RNA modeling and design, trained on OpenRNA v1 — a curated atlas of 114 million full-length RNA sequences spanning all domains of life.

omicsViewer visualizes ExpressionSet (or SummarizedExperiment) in an interactive way. The omicsViewer has a separate back- and front-end. In the back-end, users need to prepare an ExpressionSet that contains all the necessary information for the downstream data interpretation. Some extra requirements on the headers of phenotype data or feature data are imposed so that the provided information can be clearly recognized by the front-end, at the same time, keep a minimum modification on the existing ExpressionSet object. The pure dependency on R/Bioconductor guarantees maximum flexibility in the statistical analysis in the back-end. Once the ExpressionSet is prepared, it can be visualized using the front-end, implemented by shiny and plotly. Both features and samples could be selected from (data) tables or graphs (scatter plot/heatmap). Different types of analyses, such as enrichment analysis (using Bioconductor package fgsea or fisher's exact test) and STRING network analysis, will be performed on the fly and the results are visualized simultaneously. When a subset of samples and a phenotype variable is selected, a significance test on means (t-test or ranked based test; when phenotype variable is quantitative) or test of independence (chi-square or fisher’s exact test; when phenotype data is categorical) will be performed to test the association between the phenotype of interest with the selected samples. Additionally, other analyses can be easily added as extra shiny modules. Therefore, omicsViewer will greatly facilitate data exploration, many different hypotheses can be explored in a short time without the need for knowledge of R. In addition, the resulting data could be easily shared using a shiny server. Otherwise, a standalone version of omicsViewer together with designated omics data could be easily created by integrating it with portable R, which can be shared with collaborators or submitted as supplementary data together with a manuscript.

Curated paper list about LLMs for chemistry covering fine-tuning, reasoning, multi-modal models, agents, and benchmarks (COLING 2025)

ChemicalOCR is a compact vision-language model fine-tuned specifically for optical character recognition (OCR) in chemical structure images. It extracts text and bounding boxes from molecular drawings, enabling the recognition of atom labels, abbreviations, and descriptive text within chemical…

Used to determine which cell types are enriched within gene lists. The package provides tools for testing enrichments within simple gene lists (such as human disease associated genes) and those resulting from differential expression studies. The package does not depend upon any particular Single Cell Transcriptome dataset and user defined datasets can be loaded in and used in the analyses.

A C++ library for parsing and manipulating VCF files.

glmSparseNet is an R-package that generalizes sparse regression models when the features (e.g. genes) have a graph structure (e.g. protein-protein interactions), by including network-based regularizers. glmSparseNet uses the glmnet R-package, by including centrality measures of the network as penalty weights in the regularization. The current version implements regularization based on node degree, i.e. the strength and/or number of its associated edges, either by promoting hubs in the solution or orphan genes in the solution. All the glmnet distribution families are supported, namely "gaussian", "poisson", "binomial", "multinomial", "cox", and "mgaussian".

Pythonic access to FASTA files.

Create, handle, validate, visualize and convert networks in the Cytoscape exchange (CX) format to standard data types and objects. The package also provides conversion to and from objects of iGraph and graphNEL. The CX format is also used by the NDEx platform, a online commons for biological networks, and the network visualization software Cytocape.

CAGE is a widely used high throughput assay for measuring transcription start site (TSS) activity. CAGEfightR is an R/Bioconductor package for performing a wide range of common data analysis tasks for CAGE and 5'-end data in general. Core functionality includes: import of CAGE TSSs (CTSSs), tag (or unidirectional) clustering for TSS identification, bidirectional clustering for enhancer identification, annotation with transcript and gene models, correlation of TSS and enhancer expression, calculation of TSS shapes, quantification of CAGE expression as expression matrices and genome brower visualization.

R-based computational framework for a comprehensive in silico analysis of circRNAs. This computational framework allows to combine and analyze circRNAs previously detected by multiple publicly available annotation-based circRNA detection tools. It covers different aspects of circRNAs analysis from differential expression analysis, evolutionary conservation, biogenesis to functional analysis.

Github | Cite

Github | Cite

CompoundDb provides functionality to create and use (chemical) compound annotation databases from a variety of different sources such as LipidMaps, HMDB, ChEBI or MassBank. The database format allows to store in addition MS/MS spectra along with compound information. The package provides also a backend for Bioconductor's Spectra package and allows thus to match experimetal MS/MS spectra against MS/MS spectra in the database. Databases can be stored in SQLite format and are thus portable.

Skill operating layer for biomedical AI agents with 211 production-ready SKILL.md files across 7 domains (biology, pharmacology, medicine, data science, literature search), enabling modular dry-lab reasoning and protocol composition for Stanford LabOS-compatible agents

Deep learning-based variant caller

Bi-directional DNA language model based on the Mamba state space architecture, enabling efficient long-range genomic sequence modeling with linear-time complexity and built-in reverse-complement equivariance; achieves strong performance on chromatin accessibility, enhancer, and promoter prediction benchmarks (Stanford & UC Berkeley, 500+ stars)

A comprehensive package for visualizing multi-set intersections and extracting detailed subset information. VennDetail generates high-resolution visualizations including traditional Venn diagrams, Venn-pie plots, and UpSet-style plots. It provides functions to extract and combine subset details with user datasets in various formats. The package is particularly useful for bioinformatics applications but can be used for any multi-set analysis.