Skip to content

encoding.md

Latest commit

 

History

History
36 lines (28 loc) · 1.58 KB

encoding.md

File metadata and controls

36 lines (28 loc) · 1.58 KB
Raw

Graph conversion and encoded FASTA

CLI index | Source map | Graph notebook

pdbs-to-graphs

pdbs-to-graphs /path/to/pdbs /path/to/training.h5 --verbose

Uses pdbgraphmk2, not the legacy ProDy converter. The properties CSV is packaged and resolved independently of the working directory. Consult pdbs-to-graphs --help for multiprocessing controls. Inspect conversion errors and populated HDF5 groups; a nonempty file alone does not establish graph coverage. Check schema/feature compatibility with the model, particularly when reusing historical HDF5 datasets. Do not regenerate running training jobs' datasets in place.

foldcomp2fasta

foldcomp2fasta /path/to/encoder.pt /path/to/foldcomp_db \
  /path/to/encoded.fasta --device cpu --max-structures 10

The database argument is its basename, without .lookup. Remove the smoke limit only for a deliberately scoped full run. Optional --ids-file selects IDs. This streams Foldcomp structures through graph conversion and the trained encoder; it does not align them or infer a tree. Output symbols are structural tokens, not amino acids. Keep the encoder's exact character replacement convention.

See the Foldcomp notebook for chunk/batch experiments and the representation guide for geometry conversions. Both routes need a trusted checkpoint with matching graph features.