VLA-SMILES: Variable-Length-Array SMILES Descriptors in Neural Network-Based QSAR Modeling

Machine learning represents a milestone in data-driven research, including material informatics, robotics, and computer-aided drug discovery. With the continuously growing virtual synthetically available chemical space, efficient robust quantitative structure–activity relationship (QSAR) methods are required to uncover molecules with desired properties. Herein, we propose variable-length-array SMILES-based (VLA-SMILES) structural descriptors that expand conventional SMILES widely used machine learning. This representation extends family of numerically coded SMILES, particularly binary expedite discovery new deep QSAR models high predictive ability. VLA-SMILES were shown speed up training based on multilayer perceptron (MLP) optimized backpropagation (ATransformedBP), resilient propagation (iRPROP‒), Adam optimization algorithms featuring rational train–test splitting, while improving ability toward more compute-intensive format. All tested MLPs under same length-array-based showed similar convergence rate combination considered procedures. Validation Kennard–Stone splitting descriptor similarity metrics was found effective than partitioning ranking by activity biological values for entire set featured QSAR. Robustness MLP assessed via method parametric model validation. In addition, statistical H0 hypothesis testing linear regression between real observed activities F2,n−2 -criteria predictability estimation among QSAR-MLPs (with n being volume set). Both approaches validation correlate when evaluation predictabilities designed descriptors.