Robust computational design and evaluation of peptide vaccines for cellular immunity with application to SARS-CoV-2

Ge Liu, Brandon Carter, Trenton Bricken, Siddhartha Jain, Mathias Viard, Mary Carrington, David K. GiffordbioRxiv 2020 · DOI

TL;DR

This paper introduces OptiVax and EvalVax, a novel computational framework for designing and evaluating peptide vaccines. Their approach leverages machine learning and population HLA haplotype frequencies to create highly effective SARS-CoV-2 vaccines that demonstrate superior predicted population coverage compared to existing methods.

What Problem Does This Paper Address?

Developing effective peptide vaccines hinges on accurately predicting which viral peptides will be presented by a person's Major Histocompatibility Complex (MHC) molecules to elicit an immune response. A major challenge is the vast diversity of human MHC alleles (HLA types) across different populations, meaning a vaccine designed for one individual or population might not work for another. Existing methods often overlook population-specific HLA frequencies and linkage disequilibrium, leading to potentially suboptimal vaccine designs. Furthermore, ensuring vaccine peptides are conserved and unlikely to mutate is crucial for long-term efficacy.

Methods

The authors developed a two-part system:

  • OptiVax: A combinatorial machine learning approach that uses beam search to optimize peptide vaccine formulations. It aims to maximize a chosen objective function (either population coverage or expected peptide-MHC hits) by selecting a minimal set of peptides.
  • EvalVax: A robust evaluation tool that assesses vaccine candidates. It comes in two flavors:
  • EvalVax-Unlinked: Considers MHC allele frequencies independently.
  • EvalVax-Robust: Crucially, it incorporates HLA haplotype frequencies and accounts for linkage disequilibrium between HLA loci, providing a more accurate population-level coverage estimate.

Both tools employ multiple peptide-MHC binding prediction models (an ensemble approach) to increase prediction accuracy and robustness. They also incorporate filters to remove peptides predicted to be glycosylated, highly mutable, or derived from self-proteins.

Datasets

The study utilized the SARS-CoV-2 proteome for vaccine design. For population frequency data, they used:

  • dbMHC database for MHC allele frequencies.
  • Observed HLA haplotype frequencies for White, Black, and Asian populations, inferred using the Hapferret tool.
  • Experimental data on SARS-CoV-2 and SARS-CoV peptides for model validation.

Key Insights and Ideas

  • Haplotype-aware design is critical: Accounting for population-specific HLA haplotype frequencies and linkage disequilibrium significantly improves the predicted effectiveness of peptide vaccines.
  • Ensemble prediction improves robustness: Using multiple peptide-MHC binding prediction models reduces the risk of idiosyncratic errors and increases confidence in peptide display predictions.
  • Conservative binding thresholds matter: Employing a strict 50nM binding affinity threshold for peptide-MHC binding, along with filters for mutation and glycosylation, leads to more reliable and effective vaccine candidates.
  • S, M, and N proteins are key, but not sufficient alone: While these structural proteins are important sources of epitopes, designing vaccines solely from them may limit coverage for certain MHC types, suggesting the need for broader epitope coverage.

Key Results

The authors designed peptide vaccines for SARS-CoV-2 using their OptiVax framework:

  • MHC Class I: A 19-peptide vaccine achieved 93.21% predicted population coverage with at least five peptide-HLA hits per individual and 99.91% coverage with at least one hit. All peptides were predicted to be perfectly conserved.
  • MHC Class II: A 20-peptide vaccine achieved 90.17% predicted population coverage with at least five peptide-HLA hits per individual and 90.59% coverage with at least one hit. Peptides had a mutation probability < 0.001.

When evaluating 29 previously published peptide vaccine designs, their tool predicted a maximum coverage of only 58.51% for MHC class I and 71.65% for MHC class II, highlighting the superiority of their haplotype-aware approach.

Limitations and Future Work

  • Computational Uncertainty: The authors acknowledge that peptide-MHC prediction is an imperfect science and emphasize their conservative approach to mitigate model uncertainty.
  • Limited Scope of Published Vaccines: The evaluation of existing vaccines was limited to those with fewer than 150 peptides.
  • Future Work: The authors provide their OptiVax and EvalVax tools as open-source, encouraging further research and application to other diseases. They also highlight the need for early publication of vaccine designs to foster collaboration and accelerate progress.

Summary updated

Abstract

We present a combinatorial machine learning method to evaluate and optimize peptide vaccine formulations, and we find for SARS-CoV-2 that it provides superior predicted display of viral epitopes by MHC class I and MHC class II molecules over populations when compared to other candidate vaccines. Our method is robust to idiosyncratic errors in the prediction of MHC peptide display and considers target population HLA haplotype frequencies during optimization. To minimize clinical development time our methods validate vaccines with multiple peptide presentation algorithms to increase the probability that a vaccine will be effective. We optimize an objective function that is based on the presentation likelihood of a diverse set of vaccine peptides conditioned on a target population HLA haplotype distribution and expected epitope drift. We produce separate peptide formulations for MHC class I loci (HLA-A, HLA-B, and HLA-C) and class II loci (HLA-DP, HLA-DQ, and HLA-DR) to permit signal sequence based cell compartment targeting using nucleic acid based vaccine platforms. Our SARS-CoV-2 MHC class I vaccine formulations provide 93.21% predicted population coverage with at least five vaccine peptide-HLA hits on average in an individual (≥ 1 peptide 99.91%) with all vaccine peptides perfectly conserved across 4,690 geographically sampled SARS-CoV-2 genomes. Our MHC class II vaccine formulations provide 90.17% predicted coverage with at least five vaccine peptide-HLA hits on average in an individual with all peptides having observed mutation probability ≤ 0.001. We evaluate 29 previously published peptide vaccine designs with our evaluation tool with the requirement of having at least five vaccine peptide-HLA hits per individual, and they have a predicted maximum of 58.51% MHC class I coverage and 71.65% MHC class II coverage given haplotype based analysis. We provide an open source implementation of our design methods (OptiVax), vaccine evaluation tool (EvalVax), as well as the data used in our design efforts.
BibTeX
@misc{liu2020robust,
  author = {Ge Liu and Brandon Carter and Trenton Bricken and Siddhartha Jain and Mathias Viard and Mary Carrington and David K. Gifford},
  title = {{Robust computational design and evaluation of peptide vaccines for cellular immunity with application to SARS-CoV-2}},
  year = {2020},
  doi = {10.1101/2020.05.16.088989},
  howpublished = {bioRxiv},
}
Figure