A Community Challenge to Benchmark Machine Learning
Year: 2026
Type: article
Abstract: Adaptive immune receptors (B- and T-cell receptors; BCRs and TCRs) can recognise a diverse array of different antigens, initiating and maintaining the adaptive immune response. There is immense interest in academia and industry to discover immune receptors for diagnostics and therapeutics utilizing machine learning (ML). We propose the first community challenge to systematically benchmark ML methods on two tasks: given a dataset of immune state-labeled immune repertoires, predict (1) immune state (diagnostic use case), and (2) infer immune-state-associated receptors (therapeutic discovery use case). We conceived a benchmarking database of real-world relevant data including ~75,000 high-fidelity simulated (ground-truth) and experimentally generated TCR repertoires. By quantifying the performance of state-of-the-art techniques and identifying methodological gaps, this community challenge contributes to expediting ML-based solutions for immunodiagnostics and therapeutics discovery.
Source: Figshare
Authors: Kanduri, Chakravarthi, Konstantinovsky, Thomas, Rawat, Puneet, Pavlovic, Milena, May, Damon H., Elyanow, Rebecca, Howie, Bryan, Robins, Harlan S., Curca, Crina, Hariadi, Bryan, Kumar, Ashwath, Jacob, Jose, Papalexi, Efthymia, Roco, Charles, Rosenberg, Alex, Community, The AIRR, Barton, Justin, Klambauer, Gunter, Mariotti-Ferrandiz, Encarnita, Meysman, Pieter, Prak, Eline T. Luning, Cowell, Lindsay G., Brusko, Todd M., Yaari, Gur, Greiff, Victor, Kjetil Sandve, Geir
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7135035787