Yu Li

In recent years, Yu Li has conducted a series of studies aimed at advancing RNA therapeutics through AI, including work on disease and tissue modeling, RNA modeling and design, and molecular validation in the wet lab.

In disease and tissue modeling, he and the collaborative team applied a large language model-style training scheme to construct unified disease representations in a continuous, low-dimensional space. This approach enabled efficient genetic parameter estimation in the new disease space and led to the discovery of 40 genetic loci missed by all previous methods. His team also developed new methods for tissue modeling. In addition to accurately predicting cell-type fractions, the model can also infer cell-type-specific gene expression profiles with biological relevance, accelerating the precise analysis of high-throughput clinical data.

Regarding RNA modeling and design, his team developed the first general-purpose RNA foundation model (RNA-FM), trained on 23 million unannotated RNA sequences to capture the RNA sequenctial pattern and evolutionary information. He also led the team to develop DHR, an innovative method for protein homology detection that leverages protein language models and dense retrieval techniques, achieving ultrafast and highly sensitive homology searches.

Building on these sequence modeling tools, Yu led the development of RhoFold, a 3D RNA structure prediction model that won the fully automated category in CASP15 RNA structure prediction. In addition, the joint team with his group won the overall category in that competition. The updated model, RhoFold+, has outperformed AlphaFold3 in RNA 3D structure prediction tasks.

To address real-world biological challenges, these computational advances were further validated through wet-lab experiments. For example, he led the team to develop RhoDesign, and the collaborative team used it successfully to design new fluorescent Mango aptamers, 10 of which showed higher fluorescence than the natural Mango-I aptamer. RhoDesign reduced the RNA aptamer design cycle from six months to just four weeks.