Skip to main content

Browse by Domain

Jump straight to the domain tracks in the README.

Canonical domains

These 6 canonical domain slugs (5 scientific domains + cross-domain) are the domain axis used to tag every entry on the site. Use cross-domain for general-purpose resources that don't target a single field.

  • drug-discovery-chemistry — cheminformatics, retrosynthesis, docking, chemistry agents.
  • materials-science — interatomic potentials, materials APIs, discovery benchmarks.
  • genomics-biology — protein and genomic foundation models, single-cell tooling, wet-lab automation.
  • climate-earth-sciences — weather and climate models, earth observation, environmental data.
  • physics-mathematics — symbolic regression, theorem proving, scientific ML toolchains.
  • cross-domain — general-purpose entries (frameworks, surveys, foundational systems) that span fields.

See start-here for the full three-axis tagging convention.

Browse the cards

Resources by domain

Hand-picked entries grouped by canonical domain slug. Every bullet carries the full three-axis tag set (lifecycle: · domain: · type:) locked in Phase 2. Anchors below match the domain slugs verbatim so the card grid above and external links stay stable.

Drug discovery & chemistry

Cheminformatics, retrosynthesis, docking, and chemistry-aware agents. The distinctive shape here is chemistry-grounded tool use: models are wrapped with reaction templates, simulators, and safety checks so an agent can propose, score, and (sometimes) execute synthesis plans. Most live entries cluster around experiment-planning and tool-use-execution. Five to six flagships:

  • RDKit — The de-facto open-source cheminformatics toolkit; substrate for almost every Python-based chemistry agent and ML pipeline. Tags: lifecycle:tool-use-execution · domain:drug-discovery-chemistry · type:framework,tool
  • DeepChem — Open-source library for ML in chemistry, drug discovery, and materials, with a wide catalogue of featurizers, datasets, and pretrained models. Tags: lifecycle:tool-use-execution,evaluation · domain:drug-discovery-chemistry · type:framework,dataset
  • AiZynthFinder — AstraZeneca's Monte Carlo Tree Search retrosynthesis tool with neural policies; canonical reference for synthesis planning. Tags: lifecycle:experiment-planning · domain:drug-discovery-chemistry · type:framework,tool
  • ChemCrow — LLM agent equipped with 18 expert chemistry tools (synthesis, safety, search) for autonomous chemical reasoning. Tags: lifecycle:tool-use-execution · domain:drug-discovery-chemistry · type:agent-system,framework
  • DiffDock — Diffusion-generative model for blind protein–ligand docking; a strong baseline for structure-based drug discovery agents. Tags: lifecycle:tool-use-execution,evaluation · domain:drug-discovery-chemistry · type:model,paper
  • Coscientist — Boiko et al. autonomous agent (Nature 2023) that designs, plans, and executes chemistry experiments end-to-end; the canonical hypothesis-to-execution case study for the domain and the basis for its domain:drug-discovery-chemistry tag in workflows. Tags: lifecycle:hypothesis-generation,tool-use-execution · domain:drug-discovery-chemistry · type:agent-system,paper

Materials science

Interatomic potentials, materials APIs, and discovery benchmarks. The frontier is ML force fields that are accurate enough to replace expensive DFT in screening loops, paired with active-learning campaigns that propose the next composition or structure to evaluate.

  • Materials Project (mp-api) — Official Python client and REST surface for the Materials Project's computed-properties database covering hundreds of thousands of inorganic crystals; the standard data backbone for materials ML. Tags: lifecycle:literature-intelligence,experiment-planning · domain:materials-science · type:dataset,tool
  • MACE — Equivariant message-passing interatomic potential; a leading open ML force field family for molecular dynamics and screening. Tags: lifecycle:tool-use-execution · domain:materials-science · type:model,framework
  • Matbench — Standardised benchmark suite of materials property prediction tasks with leaderboards; the canonical evaluation surface for materials ML. Tags: lifecycle:evaluation · domain:materials-science · type:benchmark,dataset
  • Open Catalyst Project — Meta + CMU dataset and challenge (OC20 / OC22 / OC25) for ML-driven catalyst discovery, with millions of DFT calculations. Tags: lifecycle:experiment-planning,evaluation · domain:materials-science · type:dataset,benchmark
  • SciAgents — Multi-agent graph-reasoning framework for materials hypothesis generation across an ontological knowledge graph (Buehler). Tags: lifecycle:hypothesis-generation · domain:materials-science · type:paper,agent-system

Genomics & biology

Protein and genomic foundation models, single-cell tooling, and wet-lab automation. AI-for-bio is unusual in that structure prediction works, which makes downstream design and screening pipelines tractable; the open challenge is closing the loop with real experiments.

  • AlphaFold — DeepMind's protein structure prediction system; the canonical reference for AI-driven structural biology. Tags: lifecycle:tool-use-execution · domain:genomics-biology · type:model,framework
  • ESM (Evolutionary Scale Modeling) — Meta's protein language models (ESM-2, ESMFold) for embeddings, structure, and function prediction at scale. Tags: lifecycle:tool-use-execution · domain:genomics-biology · type:model
  • scvi-tools — Probabilistic deep generative models for single-cell omics (scRNA-seq, ATAC, spatial); the workhorse library of the scverse stack. Tags: lifecycle:tool-use-execution,evaluation · domain:genomics-biology · type:framework,model
  • Boltz-1 — MIT's open-source biomolecular structure prediction model approaching AlphaFold 3 quality, released under MIT license. Tags: lifecycle:tool-use-execution · domain:genomics-biology · type:model
  • LAB-Bench — FutureHouse benchmark for biology research tasks (literature, figures, protocols, DBQA, cloning); aimed squarely at scientific agents. Tags: lifecycle:evaluation · domain:genomics-biology · type:benchmark,dataset

Climate & earth sciences

Weather and climate models, earth observation, and environmental data. The shift here has been from numerical PDE solvers to neural surrogates trained on reanalysis data — orders of magnitude faster, competitive in accuracy, and reshaping how forecasting and downscaling get done.

  • GraphCast — DeepMind's GNN-based medium-range weather model that beats HRES on most variables at a fraction of the compute. Tags: lifecycle:tool-use-execution,evaluation · domain:climate-earth-sciences · type:model,paper
  • Pangu-Weather — Huawei's 3D Earth-Specific Transformer for global weather forecasting; first ML model to outperform NWP at all forecast lead times in 2023. Tags: lifecycle:tool-use-execution · domain:climate-earth-sciences · type:model,paper
  • Aurora — Microsoft's foundation model for atmospheric forecasting (weather, air quality, ocean waves) trained on heterogeneous earth-system data. Tags: lifecycle:tool-use-execution · domain:climate-earth-sciences · type:model
  • WeatherBench 2 — Google Research benchmark and leaderboard for global weather forecasting at 6h–10d lead times; the standard evaluation surface for ML weather models. Tags: lifecycle:evaluation · domain:climate-earth-sciences · type:benchmark,dataset
  • ClimateLearn — Open-source PyTorch library for ML-driven climate and weather modelling, with datasets, baselines, and downscaling tasks. Tags: lifecycle:experiment-planning,evaluation · domain:climate-earth-sciences · type:framework,dataset

Physics & mathematics

Symbolic regression, theorem proving, and scientific ML toolchains. This is where AI is most directly producing machine-checkable artifacts — proofs, formulas, programs — rather than just predictions, which makes evaluation unusually clean.

  • PySR — High-performance symbolic regression (Cranmer); recovers compact, interpretable equations from data and is widely used in physics ML. Tags: lifecycle:hypothesis-generation,tool-use-execution · domain:physics-mathematics · type:framework,tool
  • Lean & mathlib — The Lean 4 theorem prover and its community-built mathematics library; the substrate for most modern AI-for-math systems. Tags: lifecycle:tool-use-execution,evaluation · domain:physics-mathematics · type:framework,tool
  • DeepXDE — Library for physics-informed neural networks, operator learning, and inverse problems over PDEs and ODEs. Tags: lifecycle:tool-use-execution · domain:physics-mathematics · type:framework
  • FunSearch — DeepMind system that pairs an LLM with an evaluator to search the space of programs and discover new mathematical results. Tags: lifecycle:hypothesis-generation · domain:physics-mathematics · type:agent-system,paper
  • miniF2F — Olympiad-style formal mathematics benchmark across Lean, Metamath, and Isabelle; a standard evaluation surface for theorem-proving models. Tags: lifecycle:evaluation · domain:physics-mathematics · type:benchmark,dataset