Review Papers Claim by Claim

MHCquant2 refines immunopeptidomics tumor antigen discovery — concise critique

Main claim MHCquant2 is an open-source nf-core Nextflow pipeline that increases sensitivity for HLA class I and II peptide identification by integrating peptide property predictors (DeepLC and MS2PIP) with OpenMS and Percolator, expands benign immunopeptidome references, and refines tumor-associated antigen and neoepitope discovery MHCquant2 paper main description

Bottom line The pipeline is a robust, reproducible, and community-ready tool that meaningfully increases identifications (esp low-abundance peptides) and provides a larger benign reference for safer tumor antigen selection, but experimental and clinical validation of novel TAAs/neoepitopes remains needed Benchmark and low-abundance benefit

MHCquant2 refines immunopeptidomics tumor antigen discovery — detailed, evidence‑based review

Executive summary

MHCquant2 is presented as an nf-core Nextflow pipeline that (1) integrates retention time and fragment intensity predictors (DeepLC, MS2PIP via MS2Rescore), (2) rescoring with Percolator and OpenMS quantification modules, (3) provides timsTOF support and scalable parallelization, and (4) when applied to a novel benign dataset (n=92) and combined with public data, expands the benign immunopeptidome by >160,000 unique peptides and improves identification rates for low-abundance peptides — claims supported by deposited data and reproducible code availability Paper abstract summary and claim

What the authors did (methods snapshot)

• Built MHCquant2 as nf-core/Nextflow DSL2 pipeline using OpenMS tools, Comet search engine, Percolator rescoring, MS2Rescore to call DeepLC and MS2PIP, and OpenMS FeatureFinder quantification modules; code and containers provided for reproducibility Pipeline architecture and components
• Generated a novel benign timsTOF dataset: 92 primary benign tissue samples spanning 23 tissues, analyzed with timsTOF PASEF and processed with MHCquant2; raw data deposited to PRIDE (PXD058436) and pipeline on GitHub/Zenodo for versioned reproducibility Data deposition and cohort
• Benchmarked against FragPipe and PEAKS and re-analyzed published tumor datasets (AML, CLL, OvCa, melanoma) to evaluate tumor antigen discovery and benign overlap.

Key results (data highlights)

• MHCquant2 increased unique HLA class I binder identifications to 133,163 (13.2% higher vs FragPipe; 2.8% vs PEAKS) and HLA class II peptides to 204,291 (5.1% vs FragPipe) on the benign dataset; mean allotype improvements observed across HLA class I alleles Identification increases vs other pipelines
• MHCquant2-exclusive identifications biased to lower intensity: median intensity of pipeline-exclusive class I and II binders was 3.4-fold and 2.1-fold lower than shared peptides — indicating improved sensitivity to low-abundance peptides Low abundance enrichment
• Expanded benign reference: combining reanalyses and the novel dataset produced a benign reference of 420 HLA class I samples (213,462 unique binders) and 415 HLA class II samples (423,438 unique peptides), adding >160,000 new identifications overall; this impacted published tumor antigen lists where substantial fractions of putative TAAs were found in the expanded benign set and therefore excluded Expanded benign reference stats
• Re-analysis of prior tumor TAA sets: for AML 40.2% of published class I TAAs were excluded by the extended benign reference; CLL TAAs expanded heavily with 69.9% MHCquant2-exclusive TAAs; OvCa TAAs also changed — demonstrating that analysis pipeline and benign references materially alter candidate TAA lists Impact on published TAAs
• Neoepitope sensitivity: MHCquant2 identified additional low-abundance neoepitopes in a melanoma dataset (six additional neoepitopes with four validated by synthetic peptide spectral comparison) improving confidence in mutation‑derived peptides presentation Neoepitope detection and validation

Strengths

1. Open, reproducible, containerized, nf-core implementation enabling portability and community contributions; code, containers, and data are available (GitHub, nf‑core page, Zenodo, PRIDE) Open and reproducible resources
2. Quantitative benchmark vs two state-of-the-art pipelines across a sizable, timsTOF-derived benign dataset (n=92), demonstrating measurable gains including improved low-abundance sensitivity.
3. Careful per-run Percolator feature weight analysis showing additive contributions of DeepLC and MS2PIP to rescoring, adding mechanistic interpretability to the sensitivity gains.

Limitations, caveats, and potential biases

• Benign reference representativeness. The benign set increases coverage but is still limited by tissue sourcing (post‑mortem samples from specific centers and selected tissues); some tissue types or physiological states may remain under-sampled, which can bias exclusion of TAAs if benign sampling is incomplete Authors acknowledge benign reference and method biases
• FDR and false positives. Increased identifications remain within 1% peptide‑level FDR by Percolator, but rescoring and feature augmentation can shift score distributions — independent orthogonal validations (targeted MS, T cell reactivity, or synthetic peptide spectral matches) are required to confirm biological presentation and immunogenicity beyond spectral confidence; the authors provided some synthetic validations but large‑scale orthogonal biological validation remains pending FDR discussion and limited synthetic validation
• Platform and model generalizability. DeepLC and MS2PIP models may not be equally trained for every instrument or fragmentation regime; while authors used timsTOF-specific MS2PIP models and benchmarked across platforms, some biases (length/hydrophobicity) in pipeline-exclusive sets versus other tools suggest model-dependent effects that may not generalize universally Model and pipeline-dependent biases
• Downstream clinical significance. Removing candidate TAAs because they are seen in an expanded benign set is appropriate as a conservative safety step, but absence from benign data is necessary but not sufficient for clinical target suitability — functional immunogenicity, expression localization, and on‑target toxicity need independent evaluation.

Practical recommendations for users

1. Use MHCquant2 for reproducible high‑throughput reanalysis and when sensitive detection of low‑abundance peptides is desired; leverage the nf‑core profile and container images to keep environments consistent Reproducibility and containers
2. When generating clinical candidate lists, always cross‑validate MHCquant2 calls with orthogonal assays: synthetic peptide spectra, targeted PRM/SRM, and functional T cell assays; consider tissue expression and single‑cell data to rule out low-level healthy expression not captured by benign samples.
3. Be mindful of prediction model versions: update MS2PIP/DeepLC models for instrument type and retrain if possible for new fragmentation schemes; inspect length/hydrophobicity distributions of pipeline-exclusive peptides as sanity checks.

Open questions and next experiments

• Systematic orthogonal validation: design a tiered validation study across tissues and tumors where MHCquant2-exclusive TAAs/neoepitopes are tested by synthetic spectra, targeted MS, and T cell activation assays to quantify true positive rate and immunogenic relevance.
• Model robustness: retrain or fine-tune DeepLC and MS2PIP on instrument- and sample-specific immunopeptidomics data (including timsTOF and other ion mobility instruments) and measure identification shifts; this will clarify which gains are model-driven versus pipeline integration-driven.
• Population representation: expand benign sampling to additional tissues, age groups, ethnicities, and inflammatory states to ensure the benign reference reduces false negatives for clinically relevant TAAs.

Conclusions and confidence

Evidence supports that MHCquant2 is a meaningful, reproducible improvement to immunopeptidomics workflows: it yields measurable increases in identifications (notably low‑abundance peptides), provides a large, shareable benign reference, and materially changes candidate TAA lists from previous studies. The results are well supported by deposited data and open code; however, translation to clinical antigen selection requires orthogonal biological validation and attention to benign reference representativeness Paper conclusions

Interactive follow-ups

Plot identification comparisonsReproduce benchmarks (PXD058436)

Author pages for review and context:

Jonas Scheid | Steffen Lemke | Naomi Hoenisch-Gravel | Anna Dengler | Timo Sachsenberg | Annelies Declercq | (and others listed in paper)