Machine Learning for Plasmodium falciparum Antigenicity Prediction

Abstract

Reproducible pipeline comparing 32 machine-learning models (traditional ML, deep learning, and pretrained protein language models) for prioritizing Plasmodium falciparum protein antigen candidates on the PlasmoFAB benchmark. Includes homology-aware (MMseqs2) cross-validation, calibration, PEXEL/HT motif-baseline and ablation analyses, a proteome-wide screen of the P. falciparum 3D7 reference proteome, and a full reproducibility bundle (environment lock, checksums, split/fold/cluster identifiers, seeds, OOF predictions, and per-checkpoint proteome scores). Supports the associated manuscript and the DOME checklist.