In recent years, Li Yu has led his team in a series of research projects related to RNA therapeutics and artificial intelligence, including disease modeling, RNA sequence and structure modeling, and RNA design.
In the area of disease and tissue modeling, he and his collaborators have applied training strategies similar to large language models to construct a unified disease representation in a continuous, low-dimensional space. This allows for efficient estimation of genetic parameters in new disease spaces and has led to the discovery of 40 genetic loci previously missed by all other methods. His team also developed a tissue modeling method that not only accurately predicts cell type proportions but also infers biologically meaningful cell type-specific gene expression profiles, accelerating the precise analysis of high-throughput clinical data.
In RNA modeling and design, Li Yu’s team proposed the first general-purpose RNA foundation model in the field, trained on 23 million unannotated RNA sequences. This model extracts sequence representations and evolutionary information from RNA sequences. Additionally, his team introduced an innovative protein homology detection method called DHR, based on protein language models and dense retrieval techniques, which enables ultra-fast and highly sensitive detection of protein homologs. Leveraging these sequence modeling tools, the RNA 3D structure prediction model RhoFold, developed under his leadership, won the fully automated RNA structure prediction category in CASP15 and earned his team the overall global championship. The updated model, RhoFold+, significantly outperforms AlphaFold3 in RNA 3D structure prediction.
In addressing real-world biological problems, these findings have been further validated through wet-lab experiments. For example, Li Yu’s team developed an RNA aptamer design method called RhoDesign and used it to design new fluorescent Mango aptamers. Notably, 10 of the designed aptamers exhibited higher fluorescence than the natural Mango-I. With RhoDesign, the RNA aptamer design process can be shortened from 6 months to just 4 weeks.
MIT TR35 features top 35 innovators under 35 each year across all the fields. For more details, please refer the following page.