Maxsmi: Maximizing molecular property prediction performance with confidence estimation using SMILES augmentation and deep learning

• DOI: 10.1016/j.ailsci.2021.100014
• Publication Date: 2021-11-18T08:32:59Z
• Authors: Talia B. Kimber; Maxime Gagnebin; Andrea Volkamer
• Keywords: Lipophilicity; Data set

Accurate molecular property or activity prediction is one of the main goals in computer-aided drug design. Quantitative structure-activity relationship (QSAR) modeling and machine learning, more recently deep have become an integral part this process. Such algorithms require lots data for training which, case physico-chemical bioactivity sets, remains scarce. To address lack data, augmentation techniques are increasingly applied learning. Here, we exploit that compound can be represented by various SMILES strings as means explore several techniques. Convolutional recurrent neural networks trained on four including experimental solubility, lipophilicity, measurements. Moreover, uncertainty models assessed applying test set. Our results show improves accuracy independently learning model size data. The best strategies lead to Maxsmi models, maximize performance augmentation. findings standard deviation per correlates with associated prediction. In addition, our systematic testing different provides extensive guideline A tool using novel compounds aforementioned tasks made available at https://github.com/volkamerlab/maxsmi.