Your immune system runs a staggering computation every second: display fragments of every protein a cell makes, and let T cells decide friend from foe. Today’s digest is about teaching machines that computation — the T-cell arm of immunity, after last week’s antibodies. Reynisson et al.’s NetMHCpan predicts the presentation step, since ’the binding between MHC and antigenic peptides is the most selective step in the antigen presentation pathway.’ O’Donnell et al.’s open-source MHCflurry 2.0 adds ‘antigen processing steps that occur prior to MHC binding.’ Dash et al. showed that ‘quantifiable predictive features define epitope-specific T cell receptor repertoires’ — recognition is partly learnable. Sidhom et al.’s DeepTCR learns ‘a joint representation of a TCR by its CDR3 sequences and V/D/J gene usage.’ Montemurro et al.’s NetTCR-2.0 predicts ‘TCR-peptide binding by using paired TCRα and β sequence data’ — while warning that public data ‘overall is of low quality.’ And Lu et al.’s pMTnet targets the clinic, since ’neoantigens play a key role in the recognition of tumor cells by T cells.’ Reading the immune code.
