J. Am. Chem. Soc. | EIP-Diff：显式相互作用提示驱动的高保真 3D 分子生成框架
Ranking
Overall
82
Content
95
Popularity
50
Observed public metrics from 1 member.
Merged summary
TL;DR - EIP-Diff is an interaction-prompted diffusion framework for controllable, high-fidelity 3D molecular generation. Combining explicit residue-level constraints with experimentally resolved protein–ligand structures produced candidates with experimentally validated IDO1 inhibition as low as 0.31 nM.
- CrystalDataset contains 45,318 experimental protein–ligand complexes, reducing docking and artificial-pairing biases found in Crossdocked.
- The SE(3)-equivariant model encodes hydrogen bonds, halogen bonds, cation–π interactions, and π–π stacking as prompts during generation.
- Crystal data plus interaction prompts improved 3D ShapeSim Top-1 Dominance to 47.5%, Struct-DCS to 0.968, and molecular uniqueness to 99.9%.
- Case studies reproduced KAT6A binding geometry, optimized YTHDC1 candidates, and yielded two IDO1 inhibitors with cellular IC50 values of 0.75 and 0.31 nM.
Sources (1)
J. Am. Chem. Soc. | EIP-Diff：显式相互作用提示驱动的高保真 3D 分子生成框架
Public signals
OpenAlex citations 1
TL;DR - EIP-Diff is an interaction-prompted diffusion framework for controllable, high-fidelity 3D molecular generation. Combining explicit residue-level constraints with experimentally resolved protein–ligand structures produced candidates with experimentally validated IDO1 inhibition as low as 0.31 nM.
- CrystalDataset contains 45,318 experimental protein–ligand complexes, reducing docking and artificial-pairing biases found in Crossdocked.
- The SE(3)-equivariant model encodes hydrogen bonds, halogen bonds, cation–π interactions, and π–π stacking as prompts during generation.
- Crystal data plus interaction prompts improved 3D ShapeSim Top-1 Dominance to 47.5%, Struct-DCS to 0.968, and molecular uniqueness to 99.9%.
- Case studies reproduced KAT6A binding geometry, optimized YTHDC1 candidates, and yielded two IDO1 inhibitors with cellular IC50 values of 0.75 and 0.31 nM.