Find open-source science resources

Versatile multi-temporal geospatial foundation model for Earth observation, built on a ViT-based masked autoencoder with 3D spatiotemporal patch embeddings and geolocation/temporal metadata encoding; pretrained on 4.2M global time-series samples from NASA's Harmonized Landsat and Sentinel-2 archive at 30m resolution, with 300M/600M parameter variants and fine-tuning configs for flood detection, wildfire scar, landslide detection, crop segmentation, land cover, and biomass estimation (258+ stars, MIT License)

Equivariant graph attention Transformer (ICLR2023)

A toolbox for sparse contrastive principal component analysis (scPCA) of high-dimensional biological data. scPCA combines the stability and interpretability of sparse PCA with contrastive PCA's ability to disentangle biological signal from unwanted variation through the use of control data. Also implements and extends cPCA.

Extension of ProteinMPNN for protein sequence design in the context of small-molecule ligands, metal ions, and nucleic acids, enabling binding site engineering and co-factor redesign (Baker Lab)

Geometric deep learning model predicting transcriptional outcomes of novel single- and multi-gene perturbations using gene–gene knowledge graphs, 40% higher precision than prior methods on combinatorial perturbation prediction (Stanford, Nature Biotechnology 2024)

LLM for scientific research papers

Kolmogorov-Arnold Networks with learnable activation functions on edges instead of fixed node activations, achieving strong performance in function fitting, PDE solving, and scientific discovery with enhanced interpretability as an alternative to MLPs (MIT, 16.3K+ stars, 2024)

DegCre generates associations between differentially expressed genes (DEGs) and cis-regulatory elements (CREs) based on non-parametric concordance between differential data. The user provides GRanges of DEG TSS and CRE regions with differential p-value and optionally log-fold changes and DegCre returns an annotated Hits object with associations and their calculated probabilities. Additionally, the package provides functionality for visualization and conversion to other formats.

LLM agents across scientific domains

stJoincount facilitates the application of join count analysis to spatial transcriptomic data generated from the 10x Genomics Visium platform. This tool first converts a labeled spatial tissue map into a raster object, in which each spatial feature is represented by a pixel coded by label assignment. This process includes automatic calculation of optimal raster resolution and extent for the sample. A neighbors list is then created from the rasterized sample, in which adjacent and diagonal neighbors for each pixel are identified. After adding binary spatial weights to the neighbors list, a multi-categorical join count analysis is performed to tabulate "joins" between all possible combinations of label pairs. The function returns the observed join counts, the expected count under conditions of spatial randomness, and the variance calculated under non-free sampling. The z-score is then calculated as the difference between observed and expected counts, divided by the square root of the variance.

Large-scale biomolecular instruction dataset for chemistry/biology LLMs (ICLR2024)

Chemical language model

Provides a user-friendly interface to map on-targets and off-targets of CRISPR gRNA spacer sequences using bowtie. The alignment is fast, and can be performed using either commonly-used or custom CRISPR nucleases. The alignment can work with any reference or custom genomes. Both DNA- and RNA-targeting nucleases are supported.

The software uses the copy number segments from a text file and identifies all chromosome arms that are globally altered and computes various genome-wide scores. The following HRD scores (characteristic of BRCA-mutated cancers) are included: LST, HR-LOH, nLST and gLOH. the package is tailored for the ThermoFisher Oncoscan assay analyzed with their Chromosome Alteration Suite (ChAS) but can be adapted to any input.

Damsel provides an end to end analysis of DamID data. Damsel takes bam files from Dam-only control and fusion samples and counts the reads matching to each GATC region. edgeR is utilised to identify regions of enrichment in the fusion relative to the control. Enriched regions are combined into peaks, and are associated with nearby genes. Damsel allows for IGV style plots to be built as the results build, inspired by ggcoverage, and using the functionality and layering ability of ggplot2. Damsel also conducts gene ontology testing with bias correction through goseq, and future versions of Damsel will also incorporate motif enrichment analysis. Overall, Damsel is the first package allowing for an end to end analysis with visual capabilities. The goal of Damsel was to bring all the analysis into one place, and allow for exploratory analysis within R.

This packages simulates spatial transcriptomics data with the mean- variance relationship using a Gaussian Process model per gene.

A python package for optimizing chemical reactions using machine learning (contains 10 algorithms + several benchmarks).

A collection of microRNAs/targets from external resources, including validated microRNA-target databases (miRecords, miRTarBase and TarBase), predicted microRNA-target databases (DIANA-microT, ElMMo, MicroCosm, miRanda, miRDB, PicTar, PITA and TargetScan) and microRNA-disease/drug databases (miR2Disease, Pharmaco-miR VerSe and PhenomiR).

Batteries included genomic analysis pipeline for variant and RNA-Seq analysis, structural variant calling, annotation, and prediction.

Materials informatics benchmark

Web application for LLM-assisted manuscript review and annotation

Deep learning-based protein sequence design (inverse folding) from backbone structures, achieving 52.4% sequence recovery vs 32.9% for Rosetta, core tool in modern protein design pipelines (Baker Lab, Science 2022)

Workflow library embedded in the Go programming language, focusing on supporting complex workflow constructs, compiling to a single binary, providing powerful file naming and comprehensive audit reports for every output

Structure-aware prefix adaptation for integrating LLMs with knowledge graphs (ACM MM 2024)

Resources on ChIP-seq data which include papers, methods, links to software, and analysis.

UNIX-style FASTA manipulation tools.

Biomedical text generation

A benchmarking platform for molecular generation models.

This package provides a set of functions useful in the analysis of 3D genomic interactions. It includes the import of standard HiC data formats into R and HiC normalisation procedures. The main objective of this package is to improve the visualization and quantification of the analysis of HiC contacts through aggregation. The package allows to import 1D genomics data, such as peaks from ATACSeq, ChIPSeq, to create potential couples between features of interest under user-defined parameters such as distance between pairs of features of interest. It allows then the extraction of contact values from the HiC data for these couples and to perform Aggregated Peak Analysis (APA) for visualization, but also to compare normalized contact values between conditions. Overall the package allows to integrate 1D genomics data with 3D genomics data, providing an easy access to HiC contact values.

Client for the gypsum REST API (https://gypsum.artifactdb.com), a cloud-based file store in the ArtifactDB ecosystem. This package provides functions for uploads, downloads, and various adminstrative and management tasks. Check out the documentation at https://github.com/ArtifactDB/gypsum-worker for more details.

squallms is a Bioconductor R package that implements a "semi-labeled" approach to untargeted mass spectrometry data. It pulls in raw data from mass-spec files to calculate several metrics that are then used to label MS features in bulk as high or low quality. These metrics of peak quality are then passed to a simple logistic model that produces a fully-labeled dataset suitable for downstream analysis.

SMILES-Mapper is a small web app that allows students to easily visualize how SMILES & InChI strings are created together with other mol-block file formats such as .mol and .sdf files.

The R implementation for the Grammar of Succint Lipid Nomenclature parses different short hand notation dialects for lipid names. It normalizes them to a standard name. It further provides calculated monoisotopic masses and sum formulas for each successfully parsed lipid name and supplements it with LIPID MAPS Category and Class information. Also, the structural level and further structural details about the head group, fatty acyls and functional groups are returned, where applicable.

In silico derivatization for GC. The GC-derivatization tool converts carbonyl groups to C═N-OCH3 (MeOX) and transforms acidic protons into -Si(CH3)3 (TMS). Key functionalities include checking for specific groups, removing derivatization groups, and adding derivatization groups to molecules.

MoleculeExperiment contains functions to create and work with objects from the new MoleculeExperiment class. We introduce this class for analysing molecule-based spatial transcriptomics data (e.g., Xenium by 10X, Cosmx SMI by Nanostring, and Merscope by Vizgen). This allows researchers to analyse spatial transcriptomics data at the molecule level, and to have standardised data formats accross vendors.

The 'funOmics' package ggregates or summarizes omics data into higher level functional representations such as GO terms gene sets or KEGG metabolic pathways. The aggregated data matrix represents functional activity scores that facilitate the analysis of functional molecular sets while allowing to reduce dimensionality and provide easier and faster biological interpretations. Coordinated functional activity scores can be as informative as single molecules!

tidyFlowCore bridges the gap between flow cytometry analysis using the flowCore Bioconductor package and the tidy data principles advocated by the tidyverse. It provides a suite of dplyr-, ggplot2-, and tidyr-like verbs specifically designed for working with flowFrame and flowSet objects as if they were tibbles; however, your data remain flowCore data structures under this layer of abstraction. tidyFlowCore enables intuitive and streamlined analysis workflows that can leverage both the Bioconductor and tidyverse ecosystems for cytometry data.

Climate data benchmark for ML models

survClust is an outcome weighted integrative clustering algorithm used to classify multi-omic samples on their available time to event information. The resulting clusters are cross-validated to avoid over overfitting and output classification of samples that are molecularly distinct and clinically meaningful. It takes in binary (mutation) as well as continuous data (other omic types).

The package includes some statistical outlier detection methods for epimutations detection in DNA methylation data. The methods included in the package are MANOVA, Multivariate linear models, isolation forest, robust mahalanobis distance, quantile and beta. The methods compare a case sample with a suspected disease against a reference panel (composed of healthy individuals) to identify epimutations in the given case sample. It also contains functions to annotate and visualize the identified epimutations.

Functions to summarize DNA methylation data using regional principal components. Regional principal components are computed using principal components analysis within genomic regions to summarize the variability in methylation levels across CpGs. The number of principal components is chosen using either the Marcenko-Pasteur or Gavish-Donoho method to identify relevant signal in the data.

Ujjwal Karn.

Provides functionalities to visualize and contextualize CRISPR guide RNAs (gRNAs) on genomic tracks across nucleases and applications. Works in conjunction with the crisprBase and crisprDesign Bioconductor packages. Plots are produced using the Gviz framework.

Open language model for mathematics (7B/34B) trained on Proof-Pile-2, outperforming Minerva at equal scale on MATH benchmark, with tool use and formal theorem proving in Lean without finetuning (EleutherAI, ICLR 2024)

dinoR tests for significant differences in NOMe-seq footprints between two conditions, using genomic regions of interest (ROI) centered around a landmark, for example a transcription factor (TF) motif. This package takes NOMe-seq data (GCH methylation/protection) in the form of a Ranged Summarized Experiment as input. dinoR can be used to group sequencing fragments into 3 or 5 categories representing characteristic footprints (TF bound, nculeosome bound, open chromatin), plot the percentage of fragments in each category in a heatmap, or averaged across different ROI groups, for example, containing a common TF motif. It is designed to compare footprints between two sample groups, using edgeR's quasi-likelihood methods on the total fragment counts per ROI, sample, and footprint category.

The goal of `tpSVG` is to detect and visualize spatial variation in the gene expression for spatially resolved transcriptomics data analysis. Specifically, `tpSVG` introduces a family of count-based models, with generalizable parametric assumptions such as Poisson distribution or negative binomial distribution. In addition, comparing to currently available count-based model for spatially resolved data analysis, the `tpSVG` models improves computational time, and hence greatly improves the applicability of count-based models in SRT data analysis.

A generic C++ trie search tree library for small alphabets, allowing customizable leaf node structures and supporting approximate matching and word generation.

Predicts whether an amino acid substitution affects protein function.

A package for benchmarking of models for _de novo_ molecular design.