Neural network models for sequence-based TCR and HLA association prediction

• DOI: 10.1371/journal.pcbi.1011664
• Publication Date: 2023-11-20T18:44:10Z
• Authors: Si Liu; Philip Bradley; Wei Sun
• Keywords: QH301-705.5; HLA Antigens; Histocompatibility Antigens Class I; Receptors, Antigen, T-Cell; Histocompatibility Antigens Class II; Humans; Neural Networks, Computer; Biology (General); Article; 3. Good health; Research Article

T cells rely on their cell receptors (TCRs) to discern foreign antigens presented by human leukocyte antigen (HLA) proteins. The TCRs of an individual contain a record this individual's past immune activities, such as response infections or vaccines. Mining the TCR data may recover useful information biomarkers for related diseases conditions. Some are observed only in individuals with certain HLA alleles, and thus characterizing requires thorough understanding TCR-HLA associations. extensive diversity alleles rareness some present formidable challenge task. Existing methods either treat categorical variable represent its alphanumeric name, have limited ability generalize HLAs that not seen training process. To address challenge, we propose neural network-based method named Deep learning Prediction association (DePTH) predict associations based amino acid sequences. We demonstrate DePTH is capable making reasonable predictions associations, even when neither nor been included dataset. Furthermore, establish can be used quantify functional similarities among these associated survival outcomes cancer patients who received checkpoint blockade treatments.