International Digital Economy Academy

Antibody language model family scaling to 1.7B parameters, tokenizing sequences as overlapping tripeptides to encode local structural motifs.

Diffusion model that generates continuous-time, all-atom biomolecular trajectories, reproducing conformational kinetics far more cheaply than MD.

Protein-ligand affinity foundation model that embeds pockets and ligands in one space, unifying virtual screening with hit-to-lead optimization.

SMILES language model pretrained by editing: substructures are dropped and restored, giving fragment-level supervision for property prediction.

Antibody structure prediction returning backbone and side-chain coordinates in about a second, driven by a 650M-parameter antibody language model.

Refines the CDR loops of a predicted antibody structure with SE(3) flow matching, steered at sampling time by bond, angle and torsion potentials.