Advance drug discovery through multitasking learning techniques

A team of researchers from the Skolkovo Institute of Science and Technology, University of Vienna and Sirius University of Science and Technology published a study in Journal of Computer Aided Molecular Design Provides guidance for enhancing drug discovery through the use of multitasking learning techniques.

At universities, students often take related courses, such as physics and mathematics, which leads to a better understanding of both subjects. Likewise, learning a new language is easier for those who already have experience with languages, especially similar languages. The same principle applies to machine learning, where a neural network can better understand multiple “subjects” if it learns them simultaneously. Given that neural networks are one of the best methods for predicting the biological properties of new chemical compounds, the question arises: How can we help a neural network learn and predict the properties of chemical compounds simultaneously with respect to multiple biological targets?

The researchers analyzed three data sets for this: a data set with information on the antiviral activity of the particles and two data sets with information on the effect of the particles on various proteins in our bodies. The datasets differed in the completeness of information on each protein or virus. During the study, scientists discovered that adding data to a data set is an effective way to improve prediction accuracy. Moreover, they showed that the more informative the original dataset was, the more noticeable this improvement was. As a result of this work, the research team prepared a set of recommendations for using data enrichment technology to improve the quality and stability of predictions, as well as methods for objective evaluation of the achieved improvement.

“Multi-task learning is widely used in many scientific fields. Unsurprisingly, it is increasingly being applied to the development of new drugs. However, the potential of this approach has not yet been fully explored, presenting us with many unsolved tasks.” , said lead author of the study, Skoltech Ph.D. Candidate Ekaterina Sosnina notes. “The possibility of using multi-task learning to develop new drug candidates inspired us and we looked for ways to improve this approach. By following our recommendations, researchers in the field of drug discovery will enhance the predictive accuracy of their models and speed up the identification of new drug candidates.”