Prediksi Interaksi Senyawa-Protein Untuk Drug Repurposing Anti COVID-19 Menggunakan Deep Neural Network

Abstract

Coronavirus Disease 2019 (COVID-19) merupakan penyakit menular saluran pernapasan yang merusak sistem kesehatan di berbagai negara. Hingga saat ini belum ditemukan obat anti COVID-19 yang efektif. Drug repurposing digunakan untuk mengatasi penyakit baru selain penyakit awal obat tersebut ditujukan. Interaksi senyawa obat dan protein penyakit merupakan dasar dari drug repurposing. Pada penelitian ini dilakukan prediksi interaksi senyawa-protein COVID-19 menggunakan metode klasifikasi Deep Neural Network. Selanjutnya, dilakukan perbandingan kinerja hasil prediksi terhadap algoritme Random Forest dan K-Nearest Neighbor dengan menggunakan representasi fitur protein Amino Acid Composition dan Dipeptide Composition. Selain itu juga dibandingkan kinerja model prediksi pada dataset dengan menggunakan seleksi fitur ANOVA. Perbandingan kinerja model prediksi diukur dengan menggunakan AUROC, accuracy, precision, recall, dan F-measure. Hasil penelitian menunjukkan bahwa penggunaan PubChem fingerprint untuk representasi senyawa dan Dipeptide Composition untuk representasi protein menghasilkan nilai metrik yang paling baik dengan accuracy (0.9548), precision (0.8736), recall (0.8551), F-measure (0.8642) dan AUROC (0.9787) pada model Deep Neural Network dan seleksi fitur ANOVA.

Coronavirus Disease 2019 (COVID-19) is an infectious respiratory disease that is damaging health systems in various countries. So far, no effective anti-Covid-19 drug has been found. Drug repurposing is used to treat new diseases other than the initial disease the drug was intended for. The interaction of drug compounds and disease proteins (DTI) is the basis of drug repurposing. In this study, predictions of the COVID-19 protein-compound interactions were carried out using the Deep Neural Network classification method. Furthermore, a comparison of the performance of the prediction results against the Random Forest and K-Nearest Neighbor algorithms was carried out using Amino Acid Composition and Dipeptide Composition protein features. The performance of the prediction model is also compared using the ANOVA feature selection. The comparison of prediction is measured using AUROC, accuracy, precision, recall, and F-measure. The results use the PubChem fingerprint for compound representation, and Dipeptide Composition for protein is the best metric value ​​with accuracy (0.9548), precision (0.8736), recall (0.8551), F-measure (0.8642), and AUROC (0.9787) at Deep Neural Network model and ANOVA.