MHCnuggets 2.x
Allele-specific LSTM networks — one per allele, 148 for class I — that read peptides letter by letter, so no alignment or padding is needed, and that are trained by transfer learning from the data-richest allele.
At a glance
What it does
Handles class I and class II in one framework and scales to high-throughput neoantigen calling; for an allele with no data it starts from the network of the closest allele by supertype.
Tasks, data types and cancers
Architecture
Transfer learning in two axes: from HLA-A*02:01 to every other allele, and from affinity data to mass-spectrometry data.
Training data
IEDB binding affinities followed by fine-tuning on MS-eluted ligands; 148 class I and additional class II networks.
Linked datasets
- training IEDB — Immune Epitope Database Open download
- benchmark IEDB automated benchmark (MHC class I) Open download
Evaluation
| Benchmark / dataset | Metric | Value | External validation | Source |
|---|---|---|---|---|
| top 0.1% of a 1:999 decoy list | PPV | 0.42 as reported by the authors — computed differently from NetMHCpan's 0.83, and not the same quantity | no | Source |
How to run
pip install mhcnuggets
from mhcnuggets.src.predict import predict
predict(class_='I', peptides_path='peptides.peps', mhc='HLA-A02:01')Regulatory status and intended use
Regulatory status is quoted from the source linked above and can change. Research-use-only models must not be used for clinical decisions.
Limitations and bias
- Not pan-specific: an unseen allele is served by a neighbour's network, chosen by supertype.
- Per-allele networks inherit the extreme data imbalance directly.
Sources
This page is educational — it is not medical advice and does not replace consultation with an oncologist. Diagnostic and treatment decisions are made solely by specialist physicians.