Related Experiment Video
Updated: Jul 15, 2025
12:27
A Pipeline to Investigate the Structures and Signaling Pathways of Sphingosine 1-Phosphate Receptors
Published on: June 8, 2022
3.4K
3D-SMGE: a pipeline for scaffold-based molecular generation and evaluation
Chao Xu1, Runduo Liu2, Shuheng Huang1
1Key Laboratory of Tropical Biological Resources of Ministry of Education, School of Pharmaceutical Sciences, Hainan University, Haikou 570228, Hainan, P.R. China.
Briefings in Bioinformatics
|September 27, 2023
Summary
This study introduces 3D-SMGE, a novel deep generative model for designing 3D molecules and predicting ADMET properties. It accelerates drug discovery by generating valid, drug-like molecules from scaffolds and improving prediction accuracy.
Area of Science:
- Computational chemistry
- Drug discovery
- Artificial intelligence in medicine
Background:
- Drug discovery requires optimizing molecular structure for biological activity and ADMET properties.
- Current deep generative models struggle with efficient 3D molecular feature extraction for drug-like molecule generation.
- Existing ADMET prediction models often use single models, limiting accuracy.
Purpose of the Study:
- To develop an effective scaffold-based molecular generation and evaluation framework for drug discovery.
- To accelerate the structural optimization process in drug discovery.
- To improve the accuracy of ADMET property predictions.
Main Methods:
- Proposed 3D-SMG, a deep generative model for end-to-end 3D molecule design.
- Developed cross-aggregated continuous-filter convolution (ca-cfconv) for efficient 3D spatial feature extraction.
- Implemented a data-adaptive multi-model approach for ADMET property prediction.
Main Results:
- 3D-SMG generated valid, unique, and novel molecules with high drug-likeness.
- The multi-model ADMET prediction method achieved superior or competitive results on 24 out of 27 benchmark datasets.
- The 3D-SMGE framework effectively generates molecules from scaffolds and predicts ADMET properties.
Conclusions:
- 3D-SMGE offers a powerful tool for scaffold-based molecular generation and ADMET property prediction.
- This framework can significantly accelerate hit-to-lead structural optimizations.
- The proposed methods enhance the efficiency and accuracy of the drug discovery pipeline.
