flashtenser/OptoFlow-ML: v1.0.0 - OptoFlow-ML: MLOps Platform for Organic Photoelectronic Property Prediction

OptoFlow-ML v1.0.0 An MLOps-driven platform for automated high-throughput prediction of organic π-conjugated molecular properties, implementing the hierarchical machine learning framework described in Bhat et al. (2022, Chem. Sci., DOI: 10.1039/d2sc04676h). Key Features Production-grade model versioning: Timestamp-based immutable snapshots with provenance tracking via metadata.json One-command reproducibility: Deterministic dependency resolution through uv package manager Automated batch training: Single CLI invocation trains all 15 properties across multiple model generations Architecture-aware routing: Dynamic model selection based on model_name keywords Uncertainty quantification: Evidential regression outputs calibrated variance via Normal-Inverse-Gamma posteriors Supported Properties 15 electronic, redox, and optical properties including HOMO, LUMO, ionization potential, electron affinity, reorganization energies, and absorption/emission wavelengths. Model Hierarchy 2nd Generation: MLP baseline with RDKit descriptors + Morgan fingerprints 3rd Generation: MPNN with Set2Set readout on molecular graphs 4th Generation: MPNN + RDKit hybrid features 4th Generation Evidential: MPNN with uncertainty quantification Technical Stack Python 3.10 (DGL compatibility requirement) PyTorch 2.0.1 + DGL 1.1.2 + DGLLife 0.3.2 RDKit 2025.9.1 for molecular featurization uv for deterministic environment management Documentation Complete bilingual documentation (Chinese/English) included: Architecture design (docs/design/ARCHITECTURE.md) Setup guide (docs/guide/SETUP.md) Training workflows (docs/guide/TRAINING.md) Prediction interfaces (docs/guide/PREDICTION.md) Versioning system (docs/guide/VERSIONING.md) Citation If you use this platform in your research, please cite the original algorithmic work: Bhat, V.; Sornberger, P.; Pokuri, B. S. S.; Duke, R.; Ganapathysubramanian, B.; Risko, C. Electronic, Redox, and Optical Property Prediction of Organic π-Conjugated Molecules through a Hierarchy of Machine Learning Approaches. Chemical Science, 2022. DOI: 10.1039/d2sc04676h License MIT License