SuperPred 3.0: drug classification and target prediction-a machine learning approach.

Area of Science:

• Pharmacology
• Computational Chemistry
• Bioinformatics

Background:

• The SuperPred webserver, last updated in 2014, provides drug classification and target prediction.
• Previous versions relied on structural similarity, potentially limiting accuracy.
• Comparing prediction methods is challenging due to variations in training datasets.

Purpose of the Study:

• To present the updated SuperPred webserver (version 3.0) with enhanced drug classification and target prediction capabilities.
• To introduce thoroughly filtered datasets for Anatomical Therapeutic Chemical (ATC) and target prediction.
• To improve the accuracy and interpretability of drug property predictions.

Main Methods:

• Development of machine learning models for both ATC and target prediction, emphasizing functional groups.
• Creation of a rigorously filtered ATC dataset for reliable predictions.
• Expansion of the target prediction dataset to include non-binding substances to minimize false positives.

Main Results:

• Achieved an accuracy of 80.5% for ATC prediction, an improvement of nearly 5% over the previous version.
• Implemented a new scoring function for easily interpretable prediction values.
• Enhanced dataset filtering reduces false positives in target predictions.

Conclusions:

• SuperPred 3.0 offers state-of-the-art drug classification and target prediction.
• The use of machine learning and refined datasets significantly improves prediction accuracy.
• The webserver is publicly accessible for research and drug discovery.