data
Molecular data abstraction: Molecule, Dataset, FormatConverter
ChemAI Lab
Machine Learning & Quantum Chemistry Driven Chemical Research Platform
Python PyTorch RDKit v0.2.0 Apache 2.0
π¦ GitHub Repository
Project path
Place ChemAI Lab inside the application story
This page shows a long-term technical direction. Reviewers can return to the project section, verify certificates, or open the one-page resume.
Overview
ChemAI Lab is a unified chemistry + AI research platform designed to provide an end-to-end toolchain at the intersection of computational chemistry and machine learning. It integrates multiple open-source cheminformatics tools (RDKit, Chemprop, DeepChem, DScribe, MLatom, Molfeat) with a unified abstraction layer for data handling, feature engineering, model training, explainable AI, and quantum chemistry interfaces.
The project currently focuses on asymmetric organocatalysis with chiral phosphoric acids (CPAs), following a scientific roadmap of 5 stages across 100 development phases β from data infrastructure through to inverse molecular design.
Tech Stack
PyTorchDeep Learning
LightningTraining Pipeline
RDKitCheminformatics
scikit-learnClassic ML
NumPyNumerical
PandasData Processing
SciPyScientific
loguruLogging
ChempropOptional
DeepChemOptional
DScribeOptional
MLatomOptional
Module Architecture
16 subpackages covering the full research workflow
features
Molecular featurization: MolecularFeaturizer, FeatureStore
models
Model registry: ModelHub, ModelRegistry
nn
Neural network modules & custom layers
pipeline
Workflow orchestration: Workflow, PipelineStep
qm
Quantum chemistry interface: QMInterface
automl
Auto hyperparameter optimization
evaluation
Model evaluation metrics
xai
Explainable AI
viz
Data visualization
hub / pretrained
Model zoo & pretrained weights
cli / config / utils
CLI, config management, utilities
serialization
Serialization utilities
hub
Model hub & version management
Scientific Roadmap
Asymmetric organocatalysis with chiral phosphoric acids (CPAs) β 100 phases across 5 stages
1
Data Infrastructure
Standardized formats, descriptor libraries, reaction encoding, visualization tools
Phases 1-202
Asymmetric Catalysis Models
CPA-specific descriptors, enantioselectivity prediction, SHAP analysis, multi-task yield/ee models
Phases 21-403
Few-Shot Learning
Meta-learning, data augmentation, active learning, transfer learning
Phases 41-604
Mechanism-Driven Explainable AI
DFT feature fusion, reaction surface modeling, causal inference, physics-constrained NNs
Phases 61-805
Inverse Design
Target-driven molecular generation, inverse condition optimization, automated lab loops, multi-objective optimization
Phases 81-100Development Standards
- Code StyleRuff + Black (line length 100, target py311)
- Type CheckingMypy (non-strict)
- TestingPytest + plugins (coverage, benchmarks)
- DocumentationSphinx + mkdocs
- Git Hookspre-commit automated checks
- CIGitHub Actions
- LicenseApache 2.0