Drug sensitivity prediction across cancer types using graph isomorphism networks and biological pathway features: A

Shuang Li1,2,3,4, Quanzhong Yang1, Feifei Shen1

  • 1Institute of Medical Technology, Luoyang Polytechnic, Luoyang, Henan, China.

Plos One
|August 11, 2026
PubMed

Insights

This study introduces a novel dual-branch model combining Graph Isomorphism Network (GIN) for drug structures and Multilayer Perceptron (MLP) for pathway activities to predict drug sensitivity (IC50). The model significantly improves accuracy, outperforming existing benchmarks in precision medicine.

Area of Science:

  • Bioinformatics
  • Computational Biology
  • Pharmacogenomics

Background:

  • Drug sensitivity prediction is crucial for precision medicine.
  • Accurate estimation of IC50 values accelerates drug discovery.
  • Existing methods struggle to integrate drug 3D structures and cell line biological context.

Purpose of the Study:

  • To develop an innovative dual-branch model for enhanced drug sensitivity prediction.
  • To integrate molecular drug features with biological pathway information from cell lines.
  • To improve the accuracy of IC50 estimation in large-scale pharmacogenomics.

Main Methods:

  • Utilized Graph Isomorphism Network (GIN) for drug molecular representations.
  • Employed a Multilayer Perceptron (MLP) for 50-dimensional ssGSEA pathway activities from CCLE gene expression.
  • Trained and tested the model on the Genomics of Drug Sensitivity in Cancer 2 (GDSC2) dataset.

Main Results:

  • The proposed GIN+Pathway MLP model achieved an R2 of 0.8553 and a PCC of 0.9249 on the GDSC2 testing set.
  • Ablation studies confirmed the critical contribution of the pathway MLP component, with its removal decreasing R2 by over 0.15.
  • The model's performance surpassed established benchmarks like GraphDRP and DeepCDR on the same dataset.

Conclusions:

  • The dual-branch approach effectively integrates drug chemical structures and cell line biological context for superior drug sensitivity prediction.
  • Biological pathway information is essential for enhancing the performance of computational drug sensitivity models.
  • This approach offers a promising computational tool for accelerating precision medicine and drug discovery.