SAR and QSAR research on tyrosinase inhibitors using machine learning methods.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Tyrosinase is a critical enzyme in melanin synthesis, making it a target for treating human pigmentation disorders.
• Tyrosinase inhibitors regulate enzyme activity to control melanin production.

Purpose of the Study:

• To perform a structure-activity relationship (SAR) study on 1097 mushroom tyrosinase inhibitors.
• To develop accurate machine learning models for predicting tyrosinase inhibitor activity and classifying potential inhibitors.

Main Methods:

• Utilized five machine learning methods to build 15 classification models and 10 quantitative structure-activity relationship (QSAR) models.
• Employed ECFP4 fingerprints and RDKit descriptors with fully connected neural networks.
• Defined applicability domains using the $$ method and Williams plot for model validation.

Main Results:

• The best classification model (Model 5B) achieved 91.36% accuracy and an MCC of 0.81.
• The optimal QSAR model (Model 6J) demonstrated an r^2 of 0.770 and an RMSE of 0.482.
• Clustering identified eight distinct structural subsets of inhibitors, aiding in understanding structure-activity relationships.

Conclusions:

• Developed robust machine learning models for predicting tyrosinase inhibitor efficacy.
• The models provide valuable tools for identifying and designing novel tyrosinase inhibitors for pigmentation disorders.
• The study offers insights into the structural features governing tyrosinase inhibition.