Improved predictions of antigen presentation and TCR recognition with MixMHCpred2.2 and PRIME2.0 reveal potent SARS-CoV-2 CD8+ T-cell epitopes

Gfeller, David; Schmidt, Julien; Croce, Giancarlo; Guillaume, Philippe; Bobisse, Sara; Genolet, Raphaël; Queiroz, Lise; Cesbron, Julien; Racle, Julien; Harari, Alexandre

doi:10.1016/j.cels.2022.12.002

Cited by 50 publications

(75 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This dataset is the same as that used to evaluate NetMHCpan-4.1 6 , but with 140 deduplicated instances. Some prior methods 2,3,12,14 could not be evaluated on this dataset as they trained on the eluted ligands or do not predict set P.…”

Section: Resultsmentioning

confidence: 99%

“…BigMHC IM transfer learned on PRIME-1.0 10 and PRIME-2.0 12,14,19 datasets. The original training data consisted of viral antigens, cancer-testis antigens, neoepitopes, and 9-mer peptides randomly sampled from the human proteome to supplement the negative instances.…”

Section: Methodsmentioning

confidence: 99%

“…Mass spectrometry (MS) data, referred to as eluted ligands (EL), are naturally presented MHC ligands; MS data implicitly captures sets C, A , and B while explicitly representing set P. EL data provide positive training instances and random pMHC data are generated for negative controls. Some models 1,2,6,9–11 train on both BA and EL data, whereas other models 3,12 , including BigMHC, do not train on BA. To classify set R , a recent method 13 incorporated CDR3β sequences from the TCR to predict the binding affinity between TCR and pMHC.…”

mentioning

confidence: 99%

“…HLAthena 3 has pan-allele models that predict set P with single-layer neural networks and optionally consume transcript abundance and peptide flanking sequences. MixMHCpred 9,10,12 predicts set P , using a mixture model and position weight matrices (PWM) to extract epitope motifs. PRIME 12,14 is an extension of MixMHCpred to predict set T and was designed to infer the mechanisms of TCR recognition of pMHC complexes.…”

mentioning

confidence: 99%

“…MixMHCpred 9,10,12 predicts set P , using a mixture model and position weight matrices (PWM) to extract epitope motifs. PRIME 12,14 is an extension of MixMHCpred to predict set T and was designed to infer the mechanisms of TCR recognition of pMHC complexes.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Deep Neural Networks Predict MHC-I Epitope Presentation and Transfer Learn Neoepitope Immunogenicity

Albert

Yang

Shao

et al. 2022

Preprint

View full text Add to dashboard Cite

In this study, we present computational methods for predicting MHC-I epitope presentation with high accuracy and improved prediction of immunogenic neoepitopes. The BigMHC method comprises a large (51 million parameters), pan-allelic deep neural network trained on peptide-MHC presentation data from mass spectrometry assays and transfer learned on data from assays of antigen-specific immune response. Compared with NetMHCpan-4.1, PRIME, MHCflurry 2.0, MixMHCpred 2.1, MHCnuggets 2.3.2, and TransPHLA, BigMHC significantly improved the accuracy of predicting presented epitopes on a test set of 946,008 peptide-MHC complexes (AUROC=0.9838, AUPRC=0.9506). After transfer learning on immunogenicity data, BigMHC predicted immunogenic neoepitopes with the highest performance on independent test sets of neoepitopes validated for immunogenicity by IFN-γ release and MANAFEST assays. Moreover, BigMHC yielded significantly higher precision in identifying 215 immunogenic neoepitopes from a large (n=45,410,552) background of non-immunogenic epitopes, making BigMHC effective in clinical settings. All the data and codes are freely available at https://github.com/KarchinLab/bigmhc.

show abstract