XNeoCrypt: An Interpretable Ranking Framework for Prioritizing Noncanonical Tumor Antigen Candidates

Overview

With over 1.8 million deaths from lung cancer annually, it is imperative to develop immunotherapies that precisely target these tumors while minimizing off-target effects on healthy cells. A major bottleneck is identifying the tumor-specific antigens to target, which many current pipelines fail to solve at scale because they focus solely on DNA mutations, missing a large class of peptides caused by abnormal splicing. Recent studies have shown that these splice junctions between exons and transposable elements (JETs) can generate antigens that are highly recognizable by T cells and are specific to tumors. However, there remains no clear and interpretable way to decide which JETs are worthy of experimental validation. I engineered a novel computational framework to systematically detect and rank these JET-derived antigens in the LLC1 mouse model. The ranking framework uses interpretable, evidence-driven features such as MHC-I binding and chemical properties from each peptide to prioritize high-confidence targets for experimental validation and avoid targeting healthy cells. From over 123,487 peptides screened from a public RNA-seq dataset, I reduced the search space by over 4,900-fold to a top-25 candidate shortlist that successfully prioritized known JET candidates above random baseline across controlled evaluation trials with high statistical significance (p<0.0001). Top-ranked peptide-MHC complexes were further analyzed through structural modeling with AlphaFold to support interpretability. This turns a colossal search problem into an actionable set of targets that can move into wet-lab immunopeptidomics or T-cell assays, substantially reducing experimental burden and enabling practical noncanonical antigen discovery.

Competition history

• ISEF 2026 Computational Biology and Bioinformatics · Entry CBIO050

Resources

• ISEF abstract page ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.