Prediction of activity and selectivity profiles of human Carbonic Anhydrase inhibitors using machine learning...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Deregulation of human Carbonic Anhydrase (hCA) isoforms IX and XII is linked to various cancers.
• Selective inhibition of hCA IX and XII, sparing hCA II, is a promising strategy for developing targeted anticancer therapies with reduced side effects.
• Accurate in silico models are crucial for predicting ligand activity and selectivity against specific hCA isoforms.

Purpose of the Study:

• To develop and validate machine learning classification models for predicting the activity and selectivity of ligands against hCA isoforms II, IX, and XII.
• To improve upon traditional methods by utilizing flexible bioactivity thresholds for dataset balancing.
• To enable faster and more reliable virtual screening for potential anticancer drug candidates.

Main Methods:

• Development of machine learning classification models using high-confidence data from ChEMBL.
• Creation of training datasets with flexible bioactivity thresholds to ensure balanced active and inactive classes.
• Evaluation of multiple algorithms and sampling sizes to select high-performing models.

Main Results:

• Machine learning models demonstrated excellent performance in classifying active and inactive molecules for hCA isoforms.
• The developed models outperformed those created using conventional a priori activity threshold methods.
• The sequential application of validated models facilitates efficient and dependable virtual screening.

Conclusions:

• Validated machine learning models provide a robust tool for predicting hCA ligand activity and selectivity.
• These models enhance the efficiency and reliability of virtual screening in anticancer drug discovery.
• The approach offers a significant advancement in the development of selective hCA inhibitors for cancer therapy.