Study of COVID-19 vaccine candidate epitopes reveals low performance of common tools

New Article in Scientific Reports:
Identification and validation of 174 COVID-19 vaccine candidate epitopes reveals low performance of common epitope prediction tools

We are delighted to announce that our latest research on COVID-19 vaccine candidate epitopes has just been published in Scientific Reports. In the article, we demonstrate that common peptide-MHC prediction tools vary dramatically in performance depending on the predicted allele.

The study reveals low performance for common prediction tools when benchmarked on the SARS-CoV-2 peptides assessed in our in vitro complex-stability platform. During the study, we identified 174 peptide epitopes, some of which have been either previously or recently deposited in the IEDB, showing the cross-reactivity between SARS-CoV and SARS-CoV-2.

The study was undertaken in collaboration with researchers at University of Copenhagen (Denmark), Copenhagen University Hospital (Denmark) and Intavis (Germany).

Study Workflow

We started out by translating the SARS-CoV-2 genome sequence to proteins, and we then used NetMHC suite tools to predict potential epitopes. We validated those candidates on our in vitro peptide-MHC stability assay platform, NeoScreen. In total, we assessed 777 viral peptides over 10 HLA class I alleles and 1 HLA class II allele.

We found 174 peptides that formed a stable complex with HLA, 98 of which were previously measured and deposited in IEDB either as 9-mers or as a substring of a longer peptide, 3 peptides were reported in recent studies but not deposited in the IEDB and the 73 remaining peptides are novel. We subsequently compared the performance of 19 peptide-MHC prediction tools using our measured data.

Figure 1 depicts the study workflow.

Epitope Prediction Tools Perform Poorly

We consistently observed low performance for these tools throughout most of our dataset, with the exception from HLA-A*01:01, resulting in high false-positive rates for all tested tools. Surprisingly, we observed very low performances of the tested tools for the HLA-A*02:01 allele, which historically has been studied extensively. To investigate this shortcoming, we trained a vanilla NN on only 2,193 in-house stability measurements and found that our model outperformed all tested prediction tools in this setting.

About Immunitrack

Immunitrack is founded upon world-leading academic research on MHC-epitope binding. Using our deep knowledge in this area, we have developed the proprietary epitope screening platform technology, NeoScreen. NeoScreen accurately assays the affinity and stability of MHC/epitope interactions, and the technology has the capacity to rapidly screen libraries with thousands of (neo-) epitopes for applications within immune-oncology, vaccine production, T cell therapies and immune monitoring.