A machine learning method to predict TCR-peptide binding and the method's application in COVID-19
- T cells play a central role in viral response and vaccination.
- T cell receptor(TCR) - peptide binding provides specificity.
- By predicting TCR-peptide binding, one can "reverse engineer" the source peptides that triggered T cell response in patients.
- Vaccine design: strong response from T cells is required for maturation and proliferation of antibody-producing B cells.
- Predict binding specificity of TCR and peptides by machine learning.
- Search for SARS-CoV-2 peptides that elicit strong T cell response based on TCRs from COVID-19 patients.
- Classification Model 10-folds Acc:77.20% ± 0.78%.
- 2219 SARS-CoV-2 peptides have high binding prob (score >0.9) among COVID-19 patients.
- Excluding healthy control,16 peptides appeared in more than 6 samples.
- We proposed a supervised deep learning method to predict TCR-peptide binding.
- COVID-19 patients share more peptides with each other than with healthy individual, based on prediction given TCR sequences.
- Shared peptides across patients could be candidates for rational vaccine design.
TCR specificity (VDJdb), also, see downloaded VDJdb_TCR.tsv
10 patients TCR repertoire, also, see TCR.csv SARA-CoV-2 sequence, also, see virus_sequence.txt
Sequence homology and MHC binding ability predicted epitopes
Predictor for five of the six complementarity-determining regions (B1, B2, A1, A2 and A3) on an T-cell receptor (TCR)
Embedding six complementarity-determining regions and peptide
Transfering embedded TCR-peptide pairs to reasonable dimension
Training model for TCR-peptide pairs binary classification
Average ROC for repeated 10-fold cross validation:
