Indolizine Compound Selection for HPV Anticancer Active Prediction Using CNN Classifier with ADME Descriptors

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Machine learning in drug discovery

Background:

• Predicting structure-activity relationships (SAR) for indolizine compounds against human papilloma virus (HPV) remains challenging.
• Absorption, distribution, metabolism, excretion (ADME) descriptors are crucial for selecting effective anticancer agents.
• In silico methods are increasingly vital for accelerating drug discovery and development.

Purpose of the Study:

• To evaluate machine learning algorithms for predicting the correlation structure activity (CSA) of indolizine compounds.
• To identify the most effective algorithm for classifying indolizine compounds with potential HPV anticancer activity.
• To leverage ADME descriptors for enhanced prediction of anticancer efficacy.

Main Methods:

• Employed five machine learning algorithms: stochastic gradient descent (SGD), random forest (RF), support vector machine (SVM), convolutional neural network (CNN), and logistic regression (LR).
• Utilized ADME-related physiochemical descriptors from 8,900 indolizine compounds for classification.
• Optimized models using 26 well-established parameters and performed cross-validation analysis.

Main Results:

• The CNN model achieved the highest accuracy (98.33%) and lowest average loss (0.16).
• Other models showed competitive performance: SVM (96.03%), SGD (95.32%), LR (94.03%), and RF (93.23%).
• CNN demonstrated superior predictive capability compared to the other evaluated machine learning methods.

Conclusions:

• Machine learning, particularly CNNs, effectively predicts the anticancer activity of indolizine compounds against HPV.
• This approach facilitates early-stage prediction, aiding in the selection of promising drug candidates.
• The study highlights the utility of in silico ADME-based predictions in preclinical drug development.