NVIDIA Unveils CodonFM: A Breakthrough in RNA Design for Digital Biology
NVIDIA has launched CodonFM, an advanced RNA foundation model set to transform digital biology research. The model interprets RNA sequences with natural syntax, enabling deeper analysis of genetic codes and codon usage bias across species. Unlike conventional protein language models, CodonFM incorporates synonymous variants, improving predictions for mRNA stability and translation efficiency.
Trained on 131 million protein-coding sequences from 22,000 species, CodonFM employs a BERT-style bidirectional encoder with a context window of 6,138 ribonucleotides. This architecture captures long-range sequence patterns refined through evolution, positioning the model as a pivotal tool for mRNA design and mutation analysis.