Related Experiment Video
Updated: Jan 16, 2026
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.7K
A Multi-input Deep Learning Architecture for STAT3 Inhibitor Prediction
Kairui Liang1,2, Wenling Qin3, Yonghong Zhang1
1Chongqing Key Research Laboratory for Drug Metabolism, College of Pharmacy, Chongqing Medical University, Chongqing 400016, China.
ACS Omega
|October 6, 2025
Summary
We developed a novel machine learning model that accurately predicts Signal transducer and activator of transcription 3 (STAT3) inhibitors. This advanced model enhances prediction accuracy and interpretability for drug discovery.
Area of Science:
- Computational Chemistry
- Machine Learning
- Drug Discovery
Background:
- Signal transducer and activator of transcription 3 (STAT3) is crucial in physiological and oncogenic pathways.
- Existing machine learning models for STAT3 inhibitor prediction need improved performance and interpretability.
Purpose of the Study:
- To develop an advanced machine learning model for predicting STAT3 inhibitors.
- To enhance the predictive performance and interpretability of STAT3 inhibitor screening.
Main Methods:
- Introduced a fingerprint-enhanced graph (FPG) attention network model.
- Integrated sequence-based fingerprints and graph attention networks for feature learning.
- Utilized multilayer perceptron for molecular activity classification.
Main Results:
- The FPG model achieved the best predictive performance (AUC=0.897) among 49 tested models.
- Outperformed existing prediction models in identifying STAT3 inhibitors.
- SHAP algorithms and attention heatmaps enhanced model interpretability by revealing structure-activity relationships.
Conclusions:
- The FPG model offers superior predictive accuracy and interpretability for STAT3 inhibitor discovery.
- The developed web service (STAT3 Pro) facilitates further research and application of STAT3 inhibitor prediction.
