henhaobio
Sign in

bioRxivCancer BiologyPreprint

A Framework for Comparing Mouse Neoantigen Immunogenicity

Peter J Matulich, Carly Sprague, Victoria P. SchusteriD, Alyssa Granados, Rohan ChaudhariiD, Bree M Mohr, Megan K. RuhlandiD, Megan L. BurgeriD

PDFDOICode
In-depth full-text digest available · Jump to digest ↓

Abstract

Cytotoxic CD8+ T cell responses targeting tumor neoantigens are critical for immunotherapy efficacy and are widely studied across different preclinical mouse tumor models. Defined neoantigens are commonly identified or introduced into mouse tumors to enable tracking of tumor-specific T cell responses. However, these neoantigens differ in immunogenicity and variation in their composition may produce immune phenotypes that are mistakenly attributed to tumor-intrinsic mechanisms and complicate comparison across studies. Here, we sought to create a framework for comparing neoantigen immunogenicity by determining relative peptide-major histocompatibility complex (pMHC) affinity (KD) and stability (Koff) and in vivo neoantigen-targeted vaccine responses for 25 commonly used mouse tumor-derived and model neoantigens. We found that in silico prediction of pMHC affinity moderately stratified strong versus weak in vivo immunogenicity but could not accurately predict rank order; experimental measurements of affinity did not improve prediction. In contrast, experimental measurement of pMHC stability closely correlated with the relative magnitude of vaccine responses demonstrating strong predictive value. Thus, we report the relative stability of commonly used neoantigens as a resource and provide a simple method to benchmark novel neoantigens against this library. This tool allows contextualization of the level of immunogenicity of newly identified neoantigens and aids in comparative interpretation of tumor-immune phenotypes across studies.

In-depth summary

Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (13 Oct 2026) · CC BY 4.0

Experimental pMHC melting-temperature stability ranked in vivo CD8+ T cell responses of 25 mouse neoantigens better than affinity assays or nine in silico predictors.

Preclinical studies lack a reliable way to compare neoantigen strength across tumor models, so neoantigen-intrinsic and tumor-intrinsic effects are hard to separate. This work tests which affinity or stability measurement tracks in vivo immunogenicity and offers a two-standard benchmarking scheme with a web tool.

Key results

  1. SuggestedMelting-temperature stability of peptide-MHC complexes ranked in vivo CD8+ T cell responses to the neoantigen library well, outperforming RMA-S affinity and the in silico predictors.

    Spearman 0.81 (H2-K^b) and 0.94 (H2-D^b) between stability and percent IFNγ+ CD8+ T cells; RMA-S affinity was a worse predictor and in silico algorithms only moderate. 3 mice per group, n = 1-2 independent experiments. · pMHC Stability Correlates with In Vivo…; Figure legends

  2. ShownIn silico binding-affinity predictions ranked neoantigens poorly against measured RMA-S affinity, especially the strongest binders.

    No algorithm reached Spearman 0.6 against RMA-S KD; the best coefficient was 0.54 on either allele. NP396 was the strongest H2-D^b binder experimentally but was predicted among the weakest by most algorithms. · In Vitro Affinity Measurements Poorly Correlate…

  3. ShownReplicate affinity measurements in the public MHCflurry training data (from IEDB) disagree most for the strongest binders, which may explain poor ranking of strong neoantigens.

    Of 713,069 entries, 4,174 repeat pairs had Pearson 0.66; 30.3% differed by more than one order of magnitude and 9.6% by more than two. Matthews correlation coefficient was lowest below 50 nM and rose from ~0.4 to ~0.7 toward 500 nM. · Training Datasets Poorly Rank Strong Neoantigens; Analysis of Public Training Data

  4. ShownMeasured stability and affinity were not interchangeable, but in silico predictions tracked stability better than affinity; MHCflurry2.0 agreed best on H2-K^b.

    MHCflurry2.0 Spearman reached 0.81 versus stability for H2-K^b; no model exceeded 0.6 for H2-D^b; PickPocket1.1 was negative for H2-K^b. Melting-temperature range was 4.4°C for H2-D^b versus 14.4°C for H2-K^b. · pMHC Stability Poorly Correlates with Affinity

  5. SuggestedA two-standard scale (SIINFEKL = 100, Y5 = 0 for H2-K^b) places new neoantigens on the library's relative-strength axis; five added H2-K^b peptides fell at the low end.

    GP34 was nearest H2bc15; Trp2 and Mrps35 were about as strong as Y5. Standards for H2-D^b are NP396 and GP276. A web tool computes relative strength from three melting temperatures. · pMHC Stability Measurement to Benchmark…

Methods, data and reproducibility

Predicted MHC-I affinity for 25 neoantigens with nine algorithms and measured affinity (RMA-S assay) and stability (differential scanning fluorimetry). Spearman correlations were computed against IFNγ+ CD8+ T cell responses after dendritic cell vaccination in C57BL/6 mice.

  • Library of 19 tumor-derived and 6 model CD8+ neoantigens restricted to H2-K^b or H2-D^b · mouse (C57BL/6 MHC alleles) · n = 25 neoantigens
  • In vivo peptide-pulsed BMDC prime plus peptide boost vaccination, IFNγ+ CD8+ T cell readout · mouse (C57BL/6) · n = 3 mice per group; n = 1-2 independent experiments
  • MHCflurry2.0 training affinity data from IEDB, filtered to quantitative measurements and repeat pairs · n = 170,722 entries; 11,645 MHC-peptide combinations; 4,174 pairs
Methods & tools
NetMHCpan4.2, NetMHCpan4.1, NetMHCpan4.0, ANN4.0, SMMPMBEC 1.0, SMM1.0, PickPocket1.1, NetMHCcons1.1, MHCflurry2.0, RMA-S peptide affinity assay, Differential scanning fluorimetry (UV-exchangeable MHC monomers), Immune Epitope Database (IEDB)
Software & environment
Python 3.12.2, numpy 2.5.1, pandas 3.0.5, scipy 1.18.0, matplotlib 3.11.1; GraphPad Prism for curve fitting.
Reproducibility gaps
Flow cytometry data, in vivo vaccination data and raw melting curves are available only from the corresponding author on reasonable request. The listed URL is a web tool, not analysis code.
Limitations (5)
  • Stability does not capture antigen expression level or low-frequency subclonal variants; Rpl18 in MC38 is immunodominant despite low affinity and stability. · stated by the authors
  • High stability does not guarantee immunogenicity when affinity is very low: Cpne1 had high stability but much lower affinity and was the one mispredicted antigen of the six in a lung model. · stated by the authors
  • Other determinants such as proteasomal processing, naive T cell precursor repertoire and TCR affinity for pMHC are not modeled by the stability metric. · stated by the authors
  • In vivo ground truth rests on one vaccine format with 3 mice per group and n = 1-2 independent experiments. Per-allele correlations come from a small number of antigens, with no external validation cohort. · noted in this summary
  • Training-data noise is offered as the likely cause of poor ranking of strong antigens, but the analysis is correlative and does not test it, for example by retraining a predictor. · noted in this summary

Read this if you compare neoantigen strength across mouse tumor models and want a low-cost stability assay and reference scale instead of in silico affinity predictions.

Citations
0

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISNibeyro '22Pfitzer '22Sears '26Xia '23Buckley '20Bai '21
74% semantic matchBioinformatics
Guadalupe Nibeyro, María Romina Girotti, Laura Beatriz Prato+3 more·bioRxiv·Mar 14, 2022· 7 citations