A generative backbone design algorithm based on the ESM2 protein language model. On 24 standard design tasks, it improves the success rate by 8 percentage points and increases diversity by 33% compared to RFdiffusion.

A protein sequence design algorithm based on the Graphormer architecture. Compared to ProteinMPNN, it improves sequence diversity by 2.2 times and increases generation speed by 1.6 times.

A diffusion-based algorithm for generating dynamic protein structures. This algorithm achieves, for the first time, high-precision dynamic conformational ensemble prediction of natural random proteins (IDPs), overcoming the limitations of static structure prediction.

The team has accumulated twenty years of experience in developing its own molecular force field systems (such as BSFF2). These systems can simultaneously and accurately simulate the dynamic behavior of both structural and natural random proteins, providing high-precision energy calculation support for functional evaluation.

Using small molecule pharmacophores to "direct" peptide generation, a first in the industry.