Predict peptide–MHC binding with mhcnuggets

Calls Python's mhcnuggets.src.predict.predict inside a basilisk-managed environment to score peptides against a given MHC allele. By default the predictions are written by Python to a temporary CSV and read back in R, which avoids Python stdout capture and is typically faster and more robust.

Usage

predictMHCnuggets(
  peptides,
  allele,
  mhc_class = "I",
  output_path = NULL,
  normalize_allele = TRUE,
  model = "lstm",
  mass_spec = FALSE,
  ic50_threshold = 500L,
  max_ic50 = 50000L,
  embed_peptides = FALSE,
  binary_preds = FALSE,
  ba_models = FALSE,
  rank_output = FALSE,
  hla_env = deepmatchrEnv()
)

Arguments

peptides

Character vector of peptide sequences (one per peptide).

allele

MHC allele string. If normalize_allele=TRUE, common forms like "A0201" are normalized to "HLA-A02:01" for class I and "HLA-DRB101:01" for class II.

mhc_class

Either "I" or "II" (default "I").

output_path

Optional file path for Python to write CSV results. If NULL (default), a secure temporary file is used and deleted on exit.

normalize_allele

Logical; normalize allele to mhcnuggets' expected format (default TRUE). Set FALSE if you already use exact mhcnuggets allele names.

model

Model architecture string (default "lstm").

mass_spec

Logical; use MS-calibrated settings (default FALSE).

ic50_threshold

Numeric IC50 threshold (nM) for binding calls (default 500).

max_ic50

Numeric max IC50 (nM) for capping (default 50000).

embed_peptides

Logical; use embedding (default FALSE).

binary_preds

Logical; request binary predictions (default FALSE).

ba_models

Logical; force binding affinity models (default FALSE).

rank_output

Logical; request rank output (default FALSE).

hla_env

A basilisk environment object that contains Python + mhcnuggets (e.g., hlaFerretEnv).

Value

A data.frame. Columns depend on options:

• Default: peptide, ic50

• If binary_preds=TRUE: peptide, binary_pred (plus ic50 if emitted by model)

• If rank_output=TRUE: includes rank (0–1 or percentile)

Details

This wrapper exposes most of mhcnuggets' arguments so advanced users can fully control model choice, thresholds, and output format. It also applies a small, session-local patch so older mhcnuggets code that calls keras.optimizers.Adam(lr=...) works on newer Keras (maps lr to learning_rate).

License and Citation

mhcnuggets is licensed under the GNU General Public License v3.0. If you use MHCnuggets in your work, please cite: Shao, B., et al. (2020). MHCnuggets: A deep learning method for peptide-MHC binding prediction. bioRxiv. GitHub: https://github.com/KarchinLab/mhcnuggets

Examples

# \donttest{
# MHCnuggets requires Python/TensorFlow and is not available on Windows
if (.Platform$OS.type != "windows") {
  res <- predictMHCnuggets(
    peptides = c("SIINFEKL","LLFGYPVYV"),
    allele   = "A*02:01",
    mhc_class = "I",
    rank_output = TRUE
  )
  head(res)
}
#> Installing pyenv ...
#> Done! pyenv has been installed to '/home/runner/.local/share/r-reticulate/pyenv/bin/pyenv'.
#> Using Python: /home/runner/.pyenv/versions/3.10.19/bin/python3.10
#> Creating virtual environment '/home/runner/.cache/R/basilisk/1.22.0/deepMatchR/0.99.0/deepmatchrEnv_v2' ... 
#> + /home/runner/.pyenv/versions/3.10.19/bin/python3.10 -m venv /home/runner/.cache/R/basilisk/1.22.0/deepMatchR/0.99.0/deepmatchrEnv_v2
#> Done!
#> Installing packages: pip, wheel, setuptools
#> + /home/runner/.cache/R/basilisk/1.22.0/deepMatchR/0.99.0/deepmatchrEnv_v2/bin/python -m pip install --upgrade pip wheel setuptools
#> Installing packages: 'mhcnuggets==2.4.1', 'tensorflow==2.19.1'
#> + /home/runner/.cache/R/basilisk/1.22.0/deepMatchR/0.99.0/deepmatchrEnv_v2/bin/python -m pip install --upgrade --no-user 'mhcnuggets==2.4.1' 'tensorflow==2.19.1'
#> Virtual environment '/home/runner/.cache/R/basilisk/1.22.0/deepMatchR/0.99.0/deepmatchrEnv_v2' successfully created.
#>     peptide    ic50
#> 1  SIINFEKL 5600.06
#> 2 LLFGYPVYV  535.92
# }