Predictive models for tyrosinase inhibitors: Challenges from heterogeneous activity data determined by different...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Tyrosinase inhibitors are crucial for treating hyperpigmentation disorders.
• Quantitative Structure-Activity Relationship (QSAR) models aid in drug discovery by correlating chemical structure with biological activity.
• Heterogeneous experimental data from different protocols can significantly impact QSAR model reliability.

Purpose of the Study:

• To develop robust QSAR models for tyrosinase inhibitors.
• To investigate the impact of systematic errors in inhibitory activity data on QSAR model performance.
• To establish a method for identifying and removing outliers from heterogeneous datasets.

Main Methods:

• Quantitative Structure-Activity Relationship (QSAR) modeling using the Random Forest (RF) algorithm.
• Evaluation of model performance using out-of-bag estimation (R²OOB) and 10-fold cross-validation (Q²CV).
• Definition and application of a systematic error metric (ERRsys) to identify and remove data outliers.

Main Results:

• QSAR model performance was strongly correlated with systematic errors in the inhibitory activity data.
• Identification and removal of 13 outliers with significant systematic errors led to improved model predictability.
• A refined QSAR model achieved an R²OOB of 0.74 and a Q²CV of 0.80 after data cleaning.

Conclusions:

• Systematic errors in experimental protocols pose a significant challenge for developing reliable QSAR models.
• Defining and addressing systematic errors is essential for improving the accuracy and robustness of QSAR models, especially with heterogeneous data.
• This study provides a practical example of data curation strategies to enhance QSAR modeling for tyrosinase inhibitors.