Efficient and accurate sequence generation with small-scale protein language models

This research introduces a 'Small-Scale Protein Language Model' (SS-PLM) that uses a few million representative protein sequences and a simpler model (14.8 million parameters) to generate new, functional protein sequences. Fine-tuned on malate dehydrogenases, the SS-PLM created sequences predicted to fold correctly and be stable, performing comparably to larger models. This makes advanced AI protein design more accessible for applications like medicine and industry.