Datasets

Every dataset the bootcamp actually trains on, with a link to the session that uses it — then the catalogs and curated lists to go further.

All topics
How to read this page. The first section is what you will actually load in a notebook during the two weeks. The second is where to look when you need data for a problem the bootcamp does not cover. The third points into the per-topic pages, which list further datasets in the context of the method that uses them.

Used in this bootcamp

11 datasets
1,128 small molecules with measured log-solubility. Loaded via dc.molnet.load_delaney; the tutorial also ships derived pKa and xTB solvation columns.
MoleculeNet · used in Molecular Property Prediction
~2,050 compounds labelled for blood–brain barrier permeability. Binary classification; the canonical scaffold-split teaching set.
MoleculeNet · used in Model Training and Explaining ML Predictions
~7,800 compounds across 12 nuclear-receptor and stress-response assays. Highly imbalanced and partly unlabelled — a realistic multitask target.
MoleculeNet · used in Neural Networks in Practice
4,200 compounds with experimental logD7.4 from ChEMBL. Single-task regression baseline.
MoleculeNet · used in Explaining ML Predictions and the hackathon
DFT-computed NMR shifts over GDB10–17 molecules, used as the large pre-training corpus before fine-tuning on a small experimental shift set.
MOLDIS group · used in Few-Shot Learning
5,760 Suzuki–Miyaura couplings (7 substrate pairs × 12 ligands × 8 bases × 4 solvents) with UPLC-MS yields, from a nanomole-scale flow platform. A full combinatorial grid, so any subset can be queried offline to simulate closed-loop optimization.
Science 2018 · used in Reaction Yield Optimization
~100 reactions across four ligand sub-datasets from closed-loop flow optimization. Smaller and noisier than the Perera grid — the low-data \"hard mode\" benchmark.
React. Chem. Eng. 2016 · used in Reaction Yield Optimization
~1,536 experimentally measured conditions (ligand × base × solvent × concentration × temperature). Ships pre-computed Mordred and one-hot descriptor matrices via the EDBO codebase.
Nature 2021 · used in Reaction Yield Optimization
55 base DFT properties × 5 aggregations for a large acid/amine pool, served over an open REST API by the MolSSI Descriptor Libraries. The tutorial spends a 600-label DFT budget against it.
Digital Discovery 2024 · used in Active Learning on DFT Descriptors
Experimental protein structures, fetched per-target to build the protein–ligand docking oracle that scores generated molecules.
RCSB PDB · used in Molecular Generation
Nine ADMET endpoints on a blinded test set for the hackathon, drawn from Therapeutics Data Commons ADME tasks, the Polaris antiviral-ADMET 2025 set, and Fang et al.'s Computational-ADME release.
TDC · Polaris · used in the Day Hackathon

Collections & catalogs

15 sources
awesome-matchem-datasets
Actively maintained index of materials- and chemistry-ML datasets, grouped by modality and task. The best single starting point when you need data for a new problem.
Ben Blaiszik · GitHub
awesome-chemistry-datasets
Companion list focused on chemistry datasets and their flat-CSV mirrors — several of this bootcamp's reaction benchmarks are sourced from here.
Kevin Jablonka · GitHub
Molecules Dataset Collection
Clean CSV mirrors of the common MoleculeNet sets, ready to read straight into pandas without the DeepChem dependency. The BBBP file used in Model Training comes from here.
G. Lambard · GitHub
MoleculeNet
The standard molecular property prediction benchmark suite — solubility, toxicity, bioactivity, and quantum targets with recommended splits and baselines.
Stanford · Free
Therapeutics Data Commons (TDC)
Unified access to therapeutic ML tasks across single-instance prediction, multi-instance prediction, and generation, with standardized splits and leaderboards.
Harvard · Free
Polaris Hub
Benchmark platform emphasising rigorous, leakage-resistant evaluation for drug-discovery ML, including blinded test sets.
Polaris · Free
Open Graph Benchmark (OGB)
Large-scale graph ML benchmarks with standardized scaffold splits, including the molecular ogbg-mol* family.
Stanford · Free
Open Reaction Database (ORD)
Community-curated open repository of chemical reaction data with a strict schema, covering both HTE campaigns and literature-extracted reactions.
ORD · Free
ChEMBL
EBI's manually curated database of bioactive molecules with drug-like properties — ~2.4M compounds and their measured activities.
EMBL-EBI · Free
PubChem
The largest open chemical information repository — structures, properties, bioassays, and literature links.
NCBI · Free
ZINC
Commercially available compounds in ready-to-dock 3D formats; ZINC250k is the standard subset for generative-model training.
UCSF · Free
Materials Project
150,000+ inorganic materials with DFT-computed properties, accessible through the mp-api Python client.
LBNL · Free
MolSSI Descriptor Libraries
Open REST API serving conformer-dependent DFT descriptor sets for common reagent classes — no authentication required.
MolSSI · Free API
IUPAC Naming Benchmark (Kaggle)
Molecular structure/IUPAC-name pairs for training and evaluating name-to-structure and structure-to-name models.
Kaggle · Free
OpenML
20,000+ general-purpose datasets with programmatic access — useful for method sanity checks outside chemistry.
OpenML · Free

More datasets by topic

15 topic pages
Drug Discovery & QSAR
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
7 datasets listed
Generative Models
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
7 datasets listed
Materials Science
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
7 datasets listed
Quantum Chemistry & DFT
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
7 datasets listed
Graph Neural Networks
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
6 datasets listed
Chemical Representations
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
5 datasets listed
Self-Driving Laboratories
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
5 datasets listed
Reaction Prediction & Synthesis
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
4 datasets listed
Transformers for Chemistry
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
4 datasets listed
Active Learning
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
3 datasets listed
Bayesian Optimization
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
3 datasets listed
Deep Learning
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
3 datasets listed
LLMs & Agentic AI
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
3 datasets listed
ML Fundamentals
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
3 datasets listed
Uncertainty Quantification
Topic page with a curated Datasets section alongside its courses, tutorials, tools and papers.
3 datasets listed