#18 · ML / AI methods

AlphaFold3 and structure tools tested for TCR-peptide-MHC and antibody specificity

The dominant thread is assessment of AlphaFold 3 and related structure-based models for T cell and antibody recognition. One preprint benchmarks these models for TCR:peptide-MHC specificity and introduces a PAE Aggregator model and a pipeline reported as more than 100-fold faster. A second preprint evaluates AlphaFold3 across ubiquitinated proteins, TCR-epitope, antibody-antigen, protein-RNA and protein-lipid tasks, reporting uneven performance and dependence on training data. Adjacent published tools address downstream steps: CEDAR provides modular pipelines for generating and prioritizing candidate T cell epitopes on IEDB NextGen, SGGly predicts structure-guided N-linked glycosylation sites using ProtBERT embeddings, and TIPs applies deep learning to proteogenomic immunopeptidomics for transposable element-derived antigens. What is new is the combination of a faster pipeline and a PAE-based aggregator with explicit testing across multiple application areas.