ProtBERT-BFD

A transformer-based language model trained on 2.1 billion protein sequences to predict structural and functional biological properties.

ProtBERT-BFD leverages the BERT architecture to decode the complex language of life. Trained on the Big Fantastic Database (BFD) containing 2.1 billion sequences, this model uses 420 million parameters to generate high-dimensional vector representations of proteins. It excels at downstream tasks like secondary structure prediction and subcellular localization by capturing long-range dependencies within amino acid chains. By treating protein sequences as sentences and residues as words, ProtBERT-BFD provides a pre-trained foundation that significantly reduces the computational cost of specialized proteomics research.

https://github.com/agemaghsoudi/ProtTrans