Phase 4: Immunopeptidomics support with InfinDIA and model fine-tuning

[ypriverol]

Overview

Non-specific digestion (immunopeptidomics, HLA peptides) produces search spaces of billions of precursors. DIA-NN v2.0 introduced InfinDIA to handle this efficiently (367M precursors/minute on 64-core systems in v2.3).

InfinDIA limitations (from DIA-NN docs):

• Fully optimized only for regular DIA; reduced throughput on PASEF, DDA
• FDR control not as precise (up to 1.5-2x deviations)
• Always use second MBR pass output

Tasks

4a. Auto-detection from SDRF

When comment[cleavage agent details] contains NT=unspecific cleavage or NT=no cleavage, auto-switch to immunopeptidomics mode.

4b. Workflow Changes

New params:

Parameter	Type	Default	Description
diann_immunopeptidomics	boolean	false	Explicit override (auto-detected from SDRF)
diann_pre_select	integer	100000	InfinDIA precursor selection limit
diann_pre_select_force	boolean	true	Enforce precursor selection limit
diann_ref_library	string	null	Calibration library for InfinDIA (--ref)

When immunopeptidomics mode:

• INSILICO_LIBRARY_GENERATION: override --min-pep-len 7, --max-pep-len 25, --min-pr-charge 1, pass --pre-select, remove --missed-cleavages
• Version guard: require DIA-NN >= 2.0

4c. Model Fine-Tuning (v2.0+)

For custom PTMs not in DIA-NN's built-in training set.

New params:

Parameter	Type	Default	DIA-NN Flag
diann_tune_rt	boolean	false	--tune-rt
diann_tune_im	boolean	false	--tune-im
diann_tune_fr	boolean	false	--tune-fr (v2.3+ only)

Tuned model files emitted from INSILICO_LIBRARY_GENERATION, passed downstream via --rt-model, --im-model, --tokens.

Test Plan

• Run PXD024809 (immunopeptidomics DIA) with DIA-NN >= 2.0 — verify InfinDIA activates
• Verify unspecific cleavage auto-detected from SDRF
• Verify peptide length/charge overrides applied
• Version guard blocks InfinDIA with DIA-NN < 2.0

Test Datasets:

Dataset	Description	Instrument
PXD024809	Immunopeptidomics DIA+DDA	Orbitrap Fusion Lumos
PXD034772	Immunopeptidome lung cancer	Q Exactive HF-X