Known gaps¶
What mhctools does not do yet, and where the work is tracked. This is the one page that carries issue numbers; the rest of the documentation describes what exists. It is reviewed at release time.
Cleavage validation and coverage¶
- Cathepsin L and AEP (legumain), and cellular validation of cathepsin profiles.
Bundled ITCell B/S/H specificity profiles now score new
canonical sequences under their published assay scope. CatL/AEP remain
unavailable, and cellular/pH/exposure calibration is still a gap. Experimental observations can be imported through
reference_panels. #470 - ProsperousPlus adapter, blocked on the open-license requirement. #281
- Serum and extracellular peptidases beyond the current candidates: activated blood and inflammation-associated proteases (thrombin, plasmin, kallikrein, proteinase 3), other exopeptidases and compartment-specific enzymes. Curate activation, inhibitors, tissue exposure and assay conditions before choosing a predictor; a generic Arg/Lys or hydrophobic-residue scan would not distinguish these enzymes. #278. Native PhageScout ELANE/CTSG sequence profiles are available; independent inflammatory/whole-serum validation remains open.
- Concrete additional protease candidates, from the primary-source and artifact audit. The remaining tasks supplement the available PhageScout native sequence tracks:
| Candidate | Next step | Tracking |
|---|---|---|
| PhageScout ELANE/CTSG classifiers | Obtain exact training medians and mature-protein normalization context; keep structural models separate | #521 |
| CatL | Add assay-scoped sequence-recognition and observed-site evidence | #514 |
| AEP/legumain | Add pH-scoped Asn/Asp recognition and observed sites | #515 |
| Thrombin | Curate extended cooperative recognition alternatives | #516 |
| Plasmin | Keep Lys/Arg-conditioned assay profiles and subsite interactions separate | #517 |
| Plasma kallikrein KLKB1 | Add human plasma-enzyme evidence separately from tissue KLKs | #518 |
| ELANE/PRTN3 | Add distinct inflammatory-enzyme recognition and independent benchmarks | #519 |
PhageScout native sequence artifacts, including the optional full DESeq2
peptide tables, are reproduced. Sequence-recognition
rules can progress independently when their source scope is clear; external
accuracy and whole-matrix validation depend on the benchmark work below.
- MMP substrate predictions (CleaveNet), which need their own native endpoint
and assay validation rather than conversion to per-bond probabilities.
#476
- Remaining intracellular peptidases, including LAP3 and BLMH, and the
unsettled THOP1 epitope-substrate disagreement in the primary reports. TPP2
and NPEPPS also need exact substrate observations so their permissive rules
gain reproduction records.
#334
- Conflicting IRAP precursor sequence in Georgiadou 2010 (DIRSSVQNKL in
the text and Table I, DIRSSQVNKL in the Figure 3F caption) blocks curating
that record. #332
Benchmarks and validation¶
- Held-out, assay-specific benchmarking of the shipped cleavage models, including observed non-cleavages, terminal modifications, homologous-sequence leakage checks and abstention. Purified-enzyme turnover and disappearance of intact peptide in serum are separate endpoints. Reconstructing the Pepsickle paper's processed validation partition belongs here too. #291
- Half-life endpoints: auditing pepADMET endpoints, overlap with the Tan model, and missing local inference artifacts. #294
Downstream integration¶
Cleavage overlays, source-table reconciliation and ranking belong to other projects; mhctools provides the evidence contract.
- Topiary: combining source tables using original evidence and exposing extracellular cleavage and peptide half-life evidence.
- Vaxrank: adopting generalized Topiary source tables for vaccine construction.