NVIDIA’s Federated AI Models Advance Protein Localization Prediction
NVIDIA has pioneered a breakthrough in biological research and drug discovery through federated AI models. The company's approach, leveraging Nvidia FLARE and the BioNeMo Framework, enables precise prediction of protein subcellular localization—a critical factor in understanding cellular mechanisms and identifying therapeutic targets.
The federated learning methodology allows collaborative model training across institutions without compromising sensitive data. A tutorial demonstrates fine-tuning the ESM-2nv model for protein classification, utilizing protein sequence embeddings from datasets referenced in recent studies. Data follows biotrainer standards with FASTA-formatted sequences and ten distinct location classifications.