Accelerating Big Data Analysis through LASSO-Random Forest Algorithm in QSAR Studies.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Machine learning in bioinformatics

Background:

• Quantitative structure-activity prediction (QSAR) aims to identify novel drug-like molecules.
• Deep learning models show promise for predicting activities of large molecular datasets (Big Data).
• Challenges in deep learning for QSAR include overfitting and extensive processing.

Purpose of the Study:

• To address limitations in QSAR model development, particularly concerning efficiency and accuracy.
• To explore the effectiveness of feature selection algorithms for identifying optimal molecular descriptors.
• To investigate the performance of a LASSO-random forest model in predicting molecular activities.

Main Methods:

• Utilized the Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection to identify key molecular descriptors.
• Developed a random forest model to predict molecular activities of compounds from a Kaggle competition.
• Compared the proposed LASSO-random forest model against Boruta-random forest, deep random forest, and deep belief network models.

Main Results:

• The LASSO-random forest model demonstrated improved output correlation compared to other algorithms.
• Significantly reduced implementation time and model complexity were observed.
• Prediction accuracy was maintained while enhancing efficiency.

Conclusions:

• LASSO-random forest is an effective method for accelerating the extraction of optimal molecular descriptors from large datasets.
• This approach enhances QSAR model interpretability and prediction accuracy.
• The findings suggest a promising strategy for efficient lead compound identification in drug discovery.