Evaluating Deep Learning models for predicting ALK-5 inhibition.

Area of Science:

• * Computational chemistry and cheminformatics.
• * Machine learning applications in drug discovery.
• * Cancer therapeutics research.

Background:

• * Computational methods are integral to modern drug design.
• * Advances in machine learning and large biological databases accelerate discovery.
• * ALK-5 inhibitors are investigated as potential cancer treatments.

Purpose of the Study:

• * To compare the predictive performance of Deep Learning, Random Forest, and Support Vector Regression models.
• * To evaluate the models' ability to predict the biological activity (pIC50) of ALK-5 inhibitors.
• * To assess the generalization power of predictive models through validation.

Main Methods:

• * Development and comparison of Deep Learning, Random Forest, and Support Vector Regression models.
• * Prediction of biological activity (pIC50) for ALK-5 inhibitors.
• * Internal and external validation procedures to assess model generalization.
• * Permutation Importance analysis to determine chemical descriptor relevance.

Main Results:

• * A deep neural network model achieved the highest performance.
• * The deep neural network model demonstrated a coefficient of determination (R²) of 0.658 on the external validation set.
• * Mean Squared Error (MSE) and Mean Absolute Error (MAE) were 0.373 and 0.450, respectively.
• * Permutation Importance identified key chemical descriptors influencing biological activity prediction.

Conclusions:

• * The developed deep neural network model is suitable for predicting ALK-5 inhibitor biological activity.
• * The model can be utilized for forecasting the efficacy of novel ALK-5 inhibitors.
• * Machine learning, particularly deep learning, offers a powerful approach for drug design and discovery.