The same drug, two people, two reactions. Often, it is genetic.
Cytochrome P450 enzymes break down a large share of drugs. A mutation in the gene that codes them can change everything: the effective dose, a side effect, a treatment failure. PepFold is a pipeline that links a DNA variant to that effect on metabolism: slower, faster, or unchanged. Everything runs locally, no data leaves.
A point mutation in an enzyme's gene. The starting point.
ESMFold folds the variant protein locally, on GPU. The 3D shape.
The variant's impact on the enzyme's stability, abundance and activity.
Metabolism slower, faster, or unchanged.
The corpus is annotated from published literature and experimentally measured panels. No synthetic shortcut.
Multi-seed, reproducible. The same variant gives the same result, every run.
The pipeline is measured against experimental data, not against itself.
A prediction you can't reproduce isn't a result.
I am not a pharmacogenomics researcher: I build the pipeline and confront it with real data (published literature, measured panels). The science comes from groups in the field, never from generated data.