Developing a Semi-Supervised Approach Using a PU-Learning-Based Data Augmentation Strategy for Multitarget Drug Discovery | JoVE Visualize

Area of Science:

Computational chemistry
Machine learning in drug discovery
Medicinal chemistry

Background:

Multifactorial diseases require multitarget therapeutics, but their clinical approval is limited.
Machine learning (ML) and deep learning (DL) have advanced virtual screening in drug discovery.
Existing data augmentation methods for ML/DL models often compromise between true positive and false positive rates.

Purpose of the Study:

To investigate the combined impact of ML/DL methods, molecular representations, and data augmentation on drug discovery.
To introduce a novel semi-supervised learning framework, Negative-Augmented PU-bagging (NAPU-bagging) SVM, to address limitations in data augmentation.
To apply the developed framework for identifying multitarget-directed ligands (MTDLs).

Main Methods:

Evaluated the performance of Support Vector Machines (SVM) against state-of-the-art DL methods.
Developed and implemented the NAPU-bagging SVM, a semi-supervised learning framework using ensemble SVMs.
Trained classifiers on resampled bags containing positive, negative, and unlabeled data to manage false positive rates while maintaining high recall.
Applied the NAPU-bagging SVM to identify MTDLs, focusing on high recall for candidate compound lists.

Main Results:

SVM performance was comparable or superior to advanced DL methods in certain contexts.
NAPU-bagging SVM effectively managed false positive rates while achieving high recall.
The method identified novel MTDL candidates for ALK-EGFR and dopamine receptors with promising docking scores and binding modes.
Case studies confirmed the utility of NAPU-bagging SVM in identifying structurally novel MTDLs.

Conclusions:

The NAPU-bagging SVM framework offers a robust approach to semi-supervised learning in drug discovery.
This methodology shows significant promise for virtual screening, particularly in the challenging area of MTDL discovery.
The developed technique can improve the efficiency and success rate of identifying novel therapeutic candidates for complex diseases.