Skip to main content

Table 5 Ablation study to demonstrate the impact of discrete descriptor representation and topological graph representation for ABCD-GGNN on the ADMET prediction task

From: Topology-enhanced molecular graph representation for anti-breast cancer drug selection

Model

Dataset

Precision

Recall

F1

Discrete molecular descriptor representation (w/o)

MN

0.8942

0.8763

0.8823

HOB

0.8392

0.8550

0.8439

hERG

0.8547

0.8631

0.8561

CYP3A4

0.9274

0.9104

0.8967

Caco-2

0.8584

0.8722

0.8646

Molecular graph representation (w/o)

MN

0.7986

0.7316

0.7471

HOB

0.8006

0.8348

0.8219

hERG

0.7618

0.7092

0.7153

CYP3A4

0.8718

0.8026

0.8193

Caco-2

0.8162

0.8023

0.8025

ABCD-GGNN

MN

0.9255

0.9613

0.9430

HOB

0.8637

0.8804

0.8712

hERG

0.8914

0.8839

0.8842

CYP3A4

0.9474

0.9163

0.9355

Caco-2

0.8828

0.8832

0.8829

  1. We run all models 10 times and report the mean test precision, recall, and F1