openvax / topiary / 24422940523

88%

master: 90%

Pull #132

github

Add NetMHC-family cache loaders (part of #128)

Five new classmethods on CachedPredictor that adapt mhctools'
existing stdout parsers into cache-shaped DataFrames — no new
parsing code, just a thin wrapper layer:

- from_netmhcpan_stdout(path, mode=...) — wraps
  mhctools.parsing.parse_netmhcpan_stdout, which auto-detects
  NetMHCpan 2.8 / 3 / 4 / 4.1.  mode selects between
  binding_affinity (default) and elution_score on 4+.
- from_netmhc_stdout(path, version=...) — classic NetMHC (3 / 4 /
  4.1 via the version-specific parse_netmhc{3,4,41}_stdout).
- from_netmhcpan_cons_stdout(path) — NetMHCcons.
- from_netmhciipan_stdout(path, version=..., mode=...) —
  NetMHCIIpan legacy / 4 / 4.3.  mode selects elution_score
  (default) or binding_affinity.
- from_netmhcstabpan_stdout(path) — NetMHCstabpan (pMHC stability).

Shared helpers at module level:
- _read_text(path): file → string
- _bindings_to_dataframe(preds): list[BindingPrediction] → DataFrame
  with cache columns (peptide, allele, peptide_length, score,
  affinity, percentile_rank, value).
- _version_from_header(text): parses "NetMHCpan version 4.1b" etc.
  out of the stdout preamble.  predictor_version is auto-composed
  from that when the caller doesn't supply one.

Only caveat: mhctools parses stdout captures, not NetMHC's -xlsfile
tab-delimited output.  Documented in docs/cached.md; separate issue
if someone needs .xls support.

6 new tests with embedded sample stdout snippets (no large fixture
files in the repo).  Full suite 1099 passing (up from 1093), lint
clean, mkdocs --strict clean.

46 of 60 new or added lines in 1 file covered. (76.67%)

3268 of 3709 relevant lines covered (88.11%)

2.64 hits per line