Skip to content

Add EGAT reaction-enthalpy (DH) inference container#37

Open
bryanpiguave wants to merge 2 commits into
masterfrom
feature/egat-enthalpy-container
Open

Add EGAT reaction-enthalpy (DH) inference container#37
bryanpiguave wants to merge 2 commits into
masterfrom
feature/egat-enthalpy-container

Conversation

@bryanpiguave

Copy link
Copy Markdown
Member

Self-contained Docker/Apptainer container that predicts heat of reaction (DH, kcal/mol) from atom-mapped reaction SMILES using an embedding-based EGATReactionNet model trained on RGD1-CHNO (test MAE 1.53 kcal/mol, R2 0.983).

  • src/rxn_egat: vendored, dependency-light inference package (EGATConv, exact return_matrix/return_reactive featurization, ring detection, model).
  • predict_enthalpy_csv.py: CSV entrypoint (reaction_smiles or reactant/product columns). Both endothermic (+DH) and exothermic (-DH) predicted symmetrically.
  • Dockerfile + enthalpy.def build files; environment.yml; pytest suite.
  • CPU-optimized inference: ~8 torch threads + parallel featurization workers (~3x faster, fp32 unchanged).

Self-contained Docker/Apptainer container that predicts heat of reaction
(DH, kcal/mol) from atom-mapped reaction SMILES using an embedding-based
EGATReactionNet model trained on RGD1-CHNO (test MAE 1.53 kcal/mol, R2 0.983).

- src/rxn_egat: vendored, dependency-light inference package (EGATConv,
  exact return_matrix/return_reactive featurization, ring detection, model).
- predict_enthalpy_csv.py: CSV entrypoint (reaction_smiles or reactant/product
  columns). Both endothermic (+DH) and exothermic (-DH) predicted symmetrically.
- Dockerfile + enthalpy.def build files; environment.yml; pytest suite.
- CPU-optimized inference: ~8 torch threads + parallel featurization workers
  (~3x faster, fp32 unchanged).
@bryanpiguave bryanpiguave requested a review from erm42 June 26, 2026 11:55
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant