Treatment of coronavirus disease: Implementation of machine learning algorithms for drug screening

Background and objectives: The pandemic of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) led to the emergence of the immediate and urgent need to develop of therapeutic measure capable of reducing its impact on the health of the population and in health and economic resources. Based on the data provided by these bioassays, in this work propose the implementation of machine learning algorithms based on a Quantitative Structure-Activity Relationship (QSAR) model for drug screening of compounds with the potential to inhibit Quinone Reductase 2 (QR2) and to replace the anti-inflammatory function of chloroquine and hydroxychloroquine in the treatment of COVID-19 avoiding its adverse effects. Methods: QSAR modeling was performed to calculate the mathematical correlations between the chemical properties of QR2 inhibitor compounds, from different bioassays, and their biochemical response on QR2 activity. The values of 22 properties were obtained by means of automatic extraction techniques from PubChem's PUG REST service. The following classification algorithms were applied: Logistic Regression, Random Forest and Multi-Layer Perceptron. To perform the computational screening, 279 drugs were selected and divided into 7 groups: Group I or PubChem-Covid-19, settled for compounds labeled by PubChem as COVID-19 (n=104); Group II, drugs with structure similar to dihydroxyphenylalanine (dopa) (n=110); Group III, ubiquins (n=16); Group IV, used in clinical trials (n=18); Group V, amantadine, pramipexole, dabigatran, rotigotine and naphthoquinone (n=5); Group VI, vitamins B (n=10); and Group VII, vitamins K (n=16). A classification threshold for Active of 0.95 was established. Results: 54 compounds were identified as Actives. Camostat, relacatib, 5-Aminopyrimidine, clovamide, coenzyme Q4, decylubiquinone, sarilumab, fingolimod, rivaroxabán, prosultiamine and alinamin, for its potential use in COVID-19, were the most significant. Conclusions: It was presented a series of compounds identified by the QSAR model as QR2 inhibitors and we analyze the main drugs in that series according to their availability and current use.