Evaluating the Limitations of Protein Sequence Representations for Parkinson's Disease Classification

This project is a research artifact rather than a software product. With 0 stars and 3 forks (likely internal to the research team) and an age of only 4 days, it lacks any market traction or community adoption. Its primary value is academic: demonstrating that protein sequences alone have limited predictive power for complex multifactorial diseases like Parkinson's. From a competitive intelligence perspective, there is no moat here; the methodologies (Amino Acid Composition, ESM/ProtBERT embeddings) are standard in the bioinformatics community. The 'leakage-free evaluation' is a standard methodological requirement in ML research, not a technical innovation. It is highly susceptible to displacement as newer, more powerful protein language models (like ESM-3 or Evo) are released, which will necessitate re-running these benchmarks. Frontier labs (Meta, Google DeepMind) provide the underlying embedding models this project evaluates, but they are unlikely to pursue this specific niche classification task themselves, as it serves more as a cautionary study on the limitations of current representations.