Docking molecules into protein binding sites scored by a pluggable MLIP energy function (Meta's UMA by default, MACE-OMOL-0 and AIMNet2 also supported). Automatically preps the binding site from any PDB, then combines desolvation and ligand-strain corrections into an electronic binding energy — runnable from an AI agent.

A conversational CLI agent (powered by Ollama) that turns a plain-English request like "dock dopamine into SULT1A3" into a full docking run — PDB lookup, blind binding-site detection, and AutoDock Vina docking — with no prior knowledge of the binding site required.

An agentic system for building and training chemistry ML models (QSAR / bioactivity / property prediction). An Ollama-driven chat model chains together real, reusable tools — ChEMBL data prep, featurization, RF/LightGBM/MPNN/Chemprop training, evaluation — to satisfy a plain-language modeling request.

A full-featured command-line version of the MOdular DRug design AGent, with rich terminal output. Combines molecular and protein tools, IC50 prediction, AutoDock Vina docking, and a fine-tunable SMILES-GPT generator, with an easy node-integration system for adding new tools.

Local, no-network ADMET prediction from SMILES using two independent Chemprop v2 MPNN model families — ADMET-AI (52 endpoints + DrugBank percentiles) and Admetica (22 per-endpoint models with applicability-domain scores) — each shipped as both a batch CLI and an LLM-callable tool.

Protein structure prediction and protein/ligand cofolding with no local GPU required. OpenFold3, RosettaFold3, and ESMFold run on Modal GPUs on demand, while ESM2 embeddings run locally on CPU.

An OpenMM molecular-dynamics pipeline for solvated protein/ligand complexes and single small molecules, built on the AMBER force-field family. Prep, build, run, and analyze are separate omd subcommands so intermediates stay inspectable.

• UMADock — MLIP-scored blind docking
• dock_assist — conversational AI-assisted blind docking
• MoDrAg — modular drug-design AI agent
• MoDrAg_CLI — command-line MoDrAg
• admet_assist — local ADMET prediction
• fold — protein structure prediction and cofolding
• MD_openmm — OpenMM molecular dynamics pipeline
• Boltz — scripts for running Boltz on the API
• FAO_MOLPROP_CLI — adversarial molecule optimization from the CLI
• MolecularPropertyOptimization — agentic molecular property optimization
• sim_assist — molecular similarity measures
• GenMask — hit expansion via token unmasking
• SMILES_GPT — GPT for SMILES generation
• SMILES_VAE — variational SMILES autoencoder

• CheMLAgent — agent for cleaning CSVs, featurizing molecules, training ML models
• admet_assist — Chemprop MPNN ADMET models
• MACE_UseAndTrain — use/fine-tune MACE for small molecules
• Simple_Molecular_Graph_models — MPNN and classical ML models for molecular graphs
• smiles_embed — SMILES embedding model training
• SMILES_GPT — GPT for SMILES generation
• SMILES_VAE — variational SMILES autoencoder
• serve_azo_model — serves a trained MLP for azo dye lambda max
• InflectionLM — visualizing LLM output inflection points
• GenMask — hit expansion via token unmasking
• FAO_MOLPROP_CLI — adversarial molecule optimization
• MolecularPropertyOptimization — agentic molecular property optimization

• Gaussian16-Scripts-and-Functions — bash functions and scripts for Gaussian 16
• Many-body-input-files — Gaussian input files for 2- and 3-body interaction energies
• NBO_CorrelatedGaussians — non-Born-Oppenheimer explicitly correlated spherical floating gaussians
• ExplicitlyCorrelatedGaussians_StochasticGrowth — stochastic optimization of explicitly correlated gaussian wavefunctions
• MACE_UseAndTrain — use/fine-tune MACE from GFN2 data

• CafChem — libraries/modules for the CafChem computational chemistry / drug design tools
• CafChemTeach — notebooks for the Python, Machine Learning, and AI for Chemistry module
• CafChemQuantum — quantum computing practice and teaching code in Q#, Qiskit, and Cirq
• Java_teaching_examples — linear regression with a Java GUI