Caltech · Intensive Training Program

AI4Chemical Sciences

Applying machine learning to the chemical sciences

14 Recorded Lectures · 13 Tutorials
Slides, recordings & notebooks — free
Watch on YouTube

This bootcamp aims to give the audience an overview of the current machine learning and artificial intelligence workflows used in chemistry — showing what is currently being done, providing links to resources for students willing to learn more, and giving an introduction on how and when to integrate ML/AI workflows in a research project.

Lecture
Tutorial (hands-on)
Hackathon
Break / Free

Foundations of ML & Representations

Time
Monday
Tuesday
Wednesday
Thursday
Friday
9:00
Lecture Bootcamp Introduction Bootcamp objectives - Logistics J. Schleinitz Lecture Supervised Learning Definition, Data Splitting, Bias-variance, Examples on small datasets S. Wahler Lecture Machine-Learned Interatomic Potentials (MLIP) Graph Neural Networks, Message Passing Neural Networks (MPNN), Application to Molecular Dynamics M. Murphy Lecture Retrieval-Augmented Generation (RAG) Retrieval-augmented literature extraction, hallucination limits S. Wahler 🧑‍💻 Hackathon Day Hackathon Morning session — apply week's concepts to a chemistry challenge A. Gurajapu
☕ 10:30 – 10:45  Coffee break
10:45
Student Introductions Attendees Introduction & Areas of Research Round-table introductions, research backgrounds, interest in the bootcamp
Tutorial Model Training 101 Data preprocessing, Train/val/test splitting, Avoiding data leakage, Baseline definition S. Wahler Tutorial Machine-Learned Interatomic Potentials How to train and validate an MLIP for Molecular Dynamics using a MACE model. M. Murphy | A. Gurajapu Tutorial Chemistry RAG Assistant Index papers, Q&A on reactions S. Wahler 🧑‍💻 Hackathon Day Hackathon Late morning — team work continues
🍽️ 12:15 – 13:30  Lunch
13:30
Lecture Evolution of Molecular Representations SMILES, InChI, SELFIES, fingerprints, expert-crafted, molecular graphs J. Schleinitz Lecture Introduction to Neural Networks & Deep Learning The perceptron, Neural networks, Overview of complex architectures J. Schleinitz Lecture Learned Representations Unsupervised learning, self-supervised learning, applications including pre-training and fine-tuning J. Schleinitz Lecture Generative Models Overview of generative models including VAEs, GANs, and diffusion models C. Liu 🧑‍💻 Hackathon Day Hackathon Afternoon — final sprint & preparation of results
☕ 15:00 – 15:15  Coffee break
15:15
Tutorial Molecular Property Prediction Building a pipeline using rdkit, feature engineering, training and performance evaluation J. Schleinitz Tutorial Neural Networks in Practice MLPs on Tox21 from scratch: training loop, overfitting, and regularization J. Schleinitz Tutorial Few shot learning Transfer learning, application, failure modes... J. Schleinitz Tutorial Molecular Generation Comparison of generative models for molecular design, introduction to drug design. J. Schleinitz 🧑‍💻 Hackathon Day Hackathon Presentations & wrap-up of hackathon results

Dataset Design & LLMs

Time
Monday
Tuesday
Wednesday
Thursday
Friday
9:00
Lecture Model Interpretability Interpretable-by-design vs. post-hoc: MDI, permutation importance, ICE/PDP, LIME, and SHAP M. Yusov Lecture Design of Experiments (DOE) Factorial design, screening, response surfaces, and experimental planning strategies A. Schmidt Lecture Active Learning Query strategies: uncertainty, BALD, core-set, pool-based vs. stream-based J. Schleinitz Lecture Agentic AI for Chemistry ReAct, Coscientist, multi-agent, self-driving labs B. Reiner | J. Wade Braun 152 (room change)
Panel Your Future at the Bench What AI readiness means for the next generation of chemists
☕ 10:30 – 10:45  Coffee break
10:45
Tutorial Explaining ML Predictions Permutation importance, ICE/PDP, SHAP, and LIME on a blood–brain barrier model M. Yusov Tutorial Design of Experiments in Practice Factorial screening, response surface setup, and experiment planning workflows A. Schmidt Tutorial Active Learning 600 DFT labels, spent wisely J. Schleinitz Tutorial Does This Chemistry Task Need an LLM? Manual parser vs. LLM predictor vs. self-optimizing DSPy Flex B. Reiner | J. Wade Braun 152 (room change) 🧑‍💻 Hackathon Prepare Presentations Final polish on slides ahead of flash talks
🍽️ 12:15 – 13:30  Lunch
13:30
Lecture Data Stewardship Data management in experimental projects for efficient use in data-driven projects. A. Schmidt Lecture Bayesian Optimization Surrogates, EI/UCB/PI, BO loop, molecular & materials discovery, multi-objective BO J. Schleinitz 🧑‍💻 Hackathon Frame Your Own Project Kickoff & framing — data and split only, no models 🧑‍💻 Hackathon Frame Your Own Project Baseline, then a first model 🧑‍💻 Hackathon Flash Talks 5 slides, 5 minutes, hard stop
☕ 15:00 – 15:15  Coffee break
15:15
Tutorial Data Stewardship Organizing and managing research data, metadata standards, data sharing practices A. Schmidt Tutorial Reaction Yield Optimization with BO Suzuki coupling, BoTorch, EI vs. UCB convergence J. Schleinitz 🧑‍💻 Hackathon Frame Your Own Project Finish the one-page plan — data, model job, and the split 🧑‍💻 Hackathon Frame Your Own Project Peer review, honest evaluation, then build your 5 slides
Closing Wrap-up & Perspectives

This bootcamp exists thanks to everyone who built its content — the lectures, the hands-on tutorials, the notebooks, and the datasets behind them. Our sincere thanks to: