Molecular pKa Prediction with Deep Learning and Chemical Fingerprints

Today, drug discovery and design, the determination of molecular properties, in particular the determination of a molecule's pKa value, is essential for understanding and optimising the biological activity of drugs. In this context, in addition to traditional chemical methods, artificial intelligence techniques such as machine learning and deep learning are increasingly used to predict molecular properties and drug design processes. In this paper, we present an approach that investigates the effect of molecular properties on pKa prediction and implements this prediction using a deep learning model. The model considers molecular weight together with chemical fingerprinting methods such as Morgan fingerprinting to represent molecular structures. The dataset used in this study contains 2093 molecular data points obtained from PubChem. The method presented in the paper predicts the pKa values of many molecules with 96.66% accuracy. This can save time and money in the drug discovery, design process, and provide valuable guidance for experimental studies. The paper also presents a comprehensive analysis of the training process, accuracy metrics and performance of the deep learning model. Finally, this paper presents research that evaluates the impact of molecular features on pKa prediction and demonstrates the success of the deep learning model in these predictions