MHC Binding vs. Immunogenicity: Why IC50 Alone Doesn't Predict a T Cell Response

neoantigen immunogenicity · NetMHCpan · MHCflurry · PRIME · pMTnet · T cell response · binding prediction

The Core Problem: Binding Does Not Equal Immunogenicity

MHC binding affinity (IC50) predicts whether a peptide can physically occupy the MHC groove. It does not predict whether a T cell specific for that peptide exists in the patient's repertoire or will mount a functional response. Among peptides predicted to bind HLA-A*02:01 with IC50 < 50 nM, roughly 10–30% are actually immunogenic — measurably recognized by T cells in immunogenicity assays.

This gap is the central unsolved problem in neoantigen biology. Better immunogenicity prediction would reduce the number of candidates needed per vaccine dose, improve patient selection for neoantigen-targeted therapies, and help explain why some patients with high TMB still do not respond to checkpoint inhibitors.

Figure 1. Left: the tradeoff between MHC binding accuracy and immunogenicity prediction AUC across tools — better binding prediction does not necessarily improve immunogenicity prediction. Right: feature importance for immunogenicity prediction in a Random Forest model.

Beyond IC50: Features That Predict Immunogenicity

1. TCR contact residue foreignness

The central residues of the peptide (positions 4-6 in a 9-mer) contact the TCR directly. If these residues are similar to self-peptides in the HLA-A*02:01 peptidome, the corresponding T cell was likely deleted during thymic selection. BLOSUM62 scoring against the self-peptidome (PRIME algorithm) captures this:

from scipy.spatial.distance import cdist
import numpy as np

# Pseudocode: foreignness score
# Higher score = more foreign (different from self)
def foreignness_score(neoantigen_peptide, self_peptidome, blosum62_matrix):
    # Extract TCR contact positions (4-6 for 9-mer)
    neo_tcr_residues = neoantigen_peptide[3:6]

    # Compare against all self-peptides at same positions
    scores = []
    for self_pep in self_peptidome:
        self_tcr = self_pep[3:6]
        score = sum(blosum62_matrix[a][b] for a, b in zip(neo_tcr_residues, self_tcr))
        scores.append(score)

    # Lower similarity = higher foreignness
    return -np.max(scores)

2. HLA-peptide stability (t½)

Binding affinity (IC50) measures how well a peptide competes for MHC binding at equilibrium. Stability (half-life of the pMHC complex) measures how long it survives on the cell surface for T cell scanning. High-affinity but unstable complexes can still fail to prime T cells. NetMHCstabpan predicts pMHC half-life and is worth adding to any pipeline.

# NetMHCstabpan (standalone or via pVACseq --netmhc-stab flag)
# Returns t1/2 (hours) — longer = more stable = better T cell priming probability

# Rule of thumb: prioritize peptides with both IC50 < 50 nM AND t1/2 > 1 hour

3. Expression and clonality

As shown in Figure 1, expression (TPM) and clonality (cancer cell fraction / variant allele frequency) are the second and fourth most important features for immunogenicity after IC50. A neoantigen from a subclonal mutation in a lowly expressed gene will not generate enough pMHC complexes for T cell detection regardless of binding affinity.

Advanced Tools: Beyond NetMHCpan

HLAthena — Trained on mass spectrometry eluted ligand data (actual HLA-presented peptides, not just binding affinity). Better at predicting which peptides actually get processed and presented. Available via the Bhatt Lab web server.

PRIME (Pan-allele immunogenicity estimator) — Explicitly models TCR contact residue foreignness. Outperforms IC50-only tools in immunogenicity prediction AUC (0.76 vs. 0.68 for NetMHCpan alone in published benchmarks).

pMTnet — Deep learning model trained on paired pMHC-TCR data. Predicts TCR-pMHC binding directly rather than inferring immunogenicity from peptide features. Highest AUC in published benchmarks but requires a GPU and is computationally expensive at scale.

Practical Recommendation: An Ensemble Scoring Approach

No single tool dominates across all datasets. The most reliable approach is to combine scores:

# Ensemble scoring (pseudocode)
# Normalize each score to [0, 1] then combine

def ensemble_score(ic50, stability_t12, foreignness, expression_tpm, vaf, prime_score):
    # IC50: lower = better, normalize to 0-1
    ic50_norm = max(0, 1 - ic50 / 500)

    # Stability: higher = better
    stability_norm = min(1, stability_t12 / 4)

    # Expression: log-scale, cap at 100 TPM
    expr_norm = min(1, np.log10(expression_tpm + 1) / 2)

    # Clonality proxy
    clonality_norm = min(1, vaf * 2)

    # Weighted ensemble (weights from feature importance analysis)
    score = (0.31 * ic50_norm +
             0.18 * foreignness +     # PRIME foreignness
             0.16 * stability_norm +  # NetMHCstabpan
             0.22 * expr_norm +
             0.13 * clonality_norm)
    return score

Where the Field Is Heading

The next generation of immunogenicity prediction will incorporate:

  • Antigen processing prediction: TAP transport efficiency and proteasomal cleavage (currently in pVACseq via NetChop)
  • Patient TCR repertoire: Does the patient have naïve T cells with the right specificity? scTCR-seq of PBMCs pre-vaccination could help select which neoantigens are most likely to find a responding T cell
  • Structural modeling: AlphaFold2 + protein-protein docking for TCR-pMHC interactions is becoming computationally feasible and may add a structural layer to binding predictions

The fundamental constraint — that immunogenicity data is generated slowly through clinical trials while the number of candidate neoantigens per patient is large — means that in silico prediction will remain important even as experimental validation becomes easier.

Working with spatial transcriptomics and want to map immune infiltration patterns? See the companion blog SpatiaBio — Squidpy tutorials and spatial omics workflows.

댓글

댓글 쓰기

이 블로그의 인기 게시물

What Is a Neoantigen? A Practical Guide for Experimental Researchers

By Lociven · NeoantigenLab · July 2026 If you work in cancer immunology, you have heard the word neoantigen more times than you can count. It appears in grant proposals, immunotherapy papers, and tumor board discussions. But the explanations are usually either too shallow ("mutated peptides the immune system can recognize") or buried in computational methods papers that assume you already know what a VCF file is. This post is for the wet-lab side of the room — the researchers who understand tumor biology and immunology but want a clear picture of what neoantigens are, where they come from, and why the field cares so much about them. Figure 1. Most somatic mutations do not become immunogenic neoantigens. Each step — coding region, MHC binding, processing, T cell recognition — acts as a filter. The core idea A neoantigen is a peptide fragment, derived from a tumor-specific somatic mutation, that can be presented on the surface of a cancer cell via MHC molecule...
자세한 내용 보기

NetMHCpan Explained: How MHC Binding Prediction Works and How to Read the Output

By Lociven · NeoantigenLab · July 2026 You run NetMHCpan on your mutant peptide list and get back a table full of percentile ranks, EL scores, and BA scores. One peptide shows 0.48% rank. Another shows 2.1%. What does that mean, and which one should you follow up on? NetMHCpan is the most widely used MHC-I binding prediction tool in neoantigen research. Most papers cite it, most pipelines include it, and most wet-lab researchers who encounter its output don't fully understand what it's predicting. This post explains the model, the output, and where interpretation goes wrong. What NetMHCpan predicts NetMHCpan predicts the binding affinity between a peptide and an MHC class I molecule. More precisely, it predicts two things: Binding affinity (BA) : Estimated IC50 in nM — the concentration of peptide needed to displace 50% of a reference ligand from the HLA molecule. Lower IC50 = stronger binder. The common threshold is IC50 < 500 nM for "weak binder,...
자세한 내용 보기

Tumor Mutational Burden (TMB): What It Measures, What It Doesn't, and How to Interpret It

By Lociven · NeoantigenLab · July 2026 Your pathology report comes back with a TMB of 14 mut/Mb. The oncologist asks whether this patient is likely to respond to pembrolizumab. What do you actually say? Tumor mutational burden has become one of the most frequently cited biomarkers in solid tumor immunotherapy — and one of the most frequently misunderstood. This post explains what TMB measures, how it is calculated, what the FDA approval actually covers, and what it cannot tell you. What TMB measures — and what it doesn't TMB is a count of somatic mutations per megabase of coding sequence sequenced. That is all it is. It does not measure: Whether any of those mutations produce peptides that bind HLA Whether those peptides are actually presented on tumor cell surfaces Whether the patient has T cells that recognize them Whether the tumor microenvironment is immunologically accessible TMB is a proxy. More mutations → more potential neoantigens → higher p...
자세한 내용 보기