The system of self-consistent semi-correlations as one of the tools of cheminformatics for designing antiviral drugs

Abstract

The development of antiviral agents against SARS-CoV-2 is necessary. Specific drugs for SARS-CoV-2 are not available. In such circumstances, the computational methods of drug discovery can be an attractive addition to usual experiments for drug discovery. Here, the so-called system of semi-correlations was applied as a tool to build up predictive models for biological activity. The semi-correlation system has one traditional continuous variable and a second variable with two values: 1 for active compounds, and 0 for inactive compounds. Therefore, the semi-correlation system is a tool for building a categorical (active/inactive) model for biological activity. The semi-correlation system used to build models for providing antiviral effects for SARS-CoV-2 inhibitors has demonstrated adherence to statistical norms. For the validation set, the best model has a Matthew correlation coefficient of 0.95. Checking the predictive potential of the models built with random splits confirms that most of them exhibit quite satisfactory statistical quality.