Table 1 Average JS divergences between reference ligands and generated molecules

From: Target-aware 3D molecular generation based on guided equivariant diffusion

Datasets

JS div.

ResGen

PocketFlow

GCDM

TargetDiff

DiffSBDD

PMDM

DiffGui

PDB bind

bonds (↓)

0.576 ± 0.14

0.515 ± 0.12

0.461 ± 0.16

0.473 ± 0.17

0.469 ± 0.14

0.474 ± 0.17

0.363 ± 0.15

angles (↓)

0.312 ± 0.11

0.257 ± 0.06

0.323 ± 0.07

0.267 ± 0.11

0.334 ± 0.08

0.239 ± 0.09

0.211 ± 0.08

dihedrals (↓)

0.450 ± 0.09

0.404 ± 0.08

0.483 ± 0.14

0.372 ± 0.08

0.402 ± 0.07

0.454 ± 0.10

0.361 ± 0.10

Cross Docked

bonds (↓)

0.553 ± 0.15

0.483 ± 0.09

0.451 ± 0.17

0.467 ± 0.16

0.488 ± 0.14

0.402 ± 0.17

0.392 ± 0.18

angles (↓)

0.238 ± 0.06

0.277 ± 0.07

0.361 ± 0.11

0.258 ± 0.06

0.327 ± 0.05

0.194 ± 0.05

0.198 ± 0.05

dihedrals (↓)

0.404 ± 0.06

0.422 ± 0.08

0.490 ± 0.19

0.441 ± 0.11

0.481 ± 0.11

0.544 ± 0.13

0.423 ± 0.13

  1. JS divergence is computed on the distributions of bond, angle, and dihedral. Bold fonts indicate the best results.