A deep learning approach for the blind logP prediction in SAMPL6 challenge

Area of Science:

• Computational chemistry
• Machine learning in drug discovery

Background:

• The water-octanol partition coefficient (logP) quantifies molecular lipophilicity, a critical factor in drug discovery.
• Accurate logP prediction is essential for high-throughput screening and lead optimization.

Purpose of the Study:

• To develop and evaluate a novel computational method for predicting logP using deep neural networks.
• To assess the performance of the developed model in a blind prediction setting for the SAMPL6 challenge.

Main Methods:

• Utilized molecular fingerprints as input features for a deep neural network model.
• Trained the model on a dataset of 12,000 molecules and validated on 2,000 molecules.
• Applied the trained model for blind prediction of logP values in the SAMPL6 challenge.

Main Results:

• The deep learning model achieved a Root Mean Square Error (RMSE) of 0.61 logP units in the SAMPL6 challenge.
• The model's performance placed it in the top quarter of 92 participating submissions.
• Demonstrated the model's capability for fast, accurate, and robust high-throughput logP prediction.

Conclusions:

• Deep learning models, combined with molecular fingerprints, provide a powerful approach for predicting molecular lipophilicity.
• The developed method shows promise for accelerating drug discovery pipelines through efficient logP assessment.
• The model's performance in the SAMPL6 challenge validates its utility for real-world applications in computational chemistry.