Limitations of Sequence-Based Protein Representations for Parkinson's Disease Classification: A Leakage-Free Benchmark

In three linesComparative study of protein representations (amino acid composition, k-mers, ProtBERT, physicochemical descriptors) for Parkinson's disease classification. ProtBERT + MLP achieves F1=0.704±0.028, ROC-AUC=0.748±0.047. Primary sequences alone show limited discriminative power; structural/functional features required.

Summary generated by Claude — human-verified