sidhomj/DeepTCR

Implements both unsupervised (VAE) and supervised deep learning architectures for TCR sequence classification and repertoire analysis, with support for paired alpha/beta chains, V/D/J gene features, and HLA allotype/supertype context. Built on TensorFlow 2.0+ with GPU acceleration, featuring ensemble inference across cross-validated models, motif discovery via multinomial regression, and adaptive subsampling for large repertoires. Integrates repertoire-level and sequence-level predictions with optional Monte Carlo dropout and sparsity-regularized latent representations.

122 stars. No commits in the last 6 months. Available on PyPI.