Development and Validation of an Easily Interpretable QSAR Model for Inhibitory Activity Prediction against Dihydrofolate Reductase from Candida albicans

Author: Sharav Desai, Vijay K. Patel, Ankita S. Patel and Jaini Patel

Download PDF

Abstract

Candidiasis is a devastating infection caused by the fungi Candida albicans species of the genus Candida. The current treatment available for candidiasis is affected by drug-resistant strains. The primary objective of the study was to developa robust and accurate regression-based QSAR model to virtually predict the inhibitory activity of the dihydrofolate reductase enzyme present in Candida albicans. We have collected 281 chemical compounds with known inhibitory activity from the ChEMBL webserver. We initially used manual curation to remove blank and false entries from the downloaded databases. All the structures were converted into sdf format using OpenBabel software. We calculated more than 2400 structural descriptors for each class of chemical compound using Alvadesc software. The main challenge encountered during the study was handling such massive data produced after calculating descriptors. Several feature selection techniques are used to reduce the number of insignificant descriptors. A total of four machine learning algorithms named MLR, SVR, RF, and RT were used to build the QSAR model on the training dataset. We used R2, MAE, Y-randomization, applicability domain, and prediction reliability indicators as statistical tools to find out the robustness, stability, and predictability of the model. The model showed satisfactory results in all the calculated parameters under the acceptable range. The developed can be used to screen inhibitors against Candida albicans.

Keywords

QSAR, Candida albicans, Dihydrofolate reductase, Y-randomization, Applicability domain, prediction reliability indicator

Conclusion

To find structural significant properties for a compound to show inhibitory activity against dihydrofolate reductase enzyme, the QSAR model was developed. We used 281 compounds with known and proven IC50 values for the model development. Feature selection and supervised machine learning techniques were used to develop the model. The statistical validation results obtained showed good predictive ability based on both internal and external parameters for validation. The model developed can be used to screen a large pool of compounds for their inhibitory activity and features obtained through studies can be used to design novel inhibitors. The list of the inhibitors screened from the virtual screening can be used for the further development of novel inhibitors.

References

-

How to cite this article

Sharav Desai, Vijay K. Patel, Ankita S. Patel and Jaini Patel (2023). Development and Validation of an Easily Interpretable QSAR Model for Inhibitory Activity Prediction against Dihydrofolate Reductase from Candida albicans. Biological Forum – An International Journal, 15(1): 505-513.