Drug Discovery for Periodontitis Treatment Based on Big Data Mining, Systems Biology, and Deep Learning Methods

Abstract

Periodontitis, a chronic inflammatory oral condition triggered by bacteria, archaea, viruses, and eukaryotic organisms, is a well-known and widespread disease around the world. While there are effective treatments for periodontitis, there are also several shortcomings associated with its management, including limited treatment options, the risk of recurrence, and the high cost of treatment. Our goal is to develop a more efficient, systematic drug design for periodontitis before clinical trials. We work on systems drug discovery and design for periodontitis treatment via systems biology and deep learning methods. We first applied big database mining to build a candidate genome-wide genetic and epigenetic network (GWGEN), which includes a protein-protein interaction network (PPIN) and a gene regulatory network (GRN) for periodontitis and healthy control. Next, based on the unhealthy and healthy microarray data, we applied system identification and system order detection methods to remove false positives in candidate GWGENs to obtain real GWGENs for periodontitis and healthy control, respectively. After the real GWGENs were obtained, we picked out the core GWGENs based on how significant the proteins and genes were via the principal network projection (PNP) method. Finally, referring to the annotation of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, we built up the core signaling pathways of periodontitis and healthy control. Consequently, we investigated the pathogenic mechanism of periodontitis by comparing their core signaling pathways. By checking up on the downstream core signaling pathway and the corresponding cellular dysfunctions of periodontitis, we identified the fos proto-oncogene, AP-1 Transcription Factor Subunit (FOS), TSC Complex Subunit 2 (TSC2), Forkhead Box O1 (FOXO1), and nuclear factor kappa-light chain enhancer of activated B cells (NF-κB) as significant biomarkers on which we could find candidate molecular drugs to target. To achieve our ultimate goal of designing a combination of molecular drugs for periodontitis treatment, a deep neural network (DNN)-based drug-target interaction (DTI) model was employed. The model is trained with the existing drug-target interaction databases for the prediction of candidate molecular drugs for significant biomarkers. Finally, we filter out brucine, disulfiram, verapamil, and PK-11195 as potential molecular drugs to be combined as a multiple-molecular drug to target the significant biomarkers based on drug design specifications, i.e., adequate drug regulation ability, high sensitivity, and low toxicity. In conclusion, we investigated the pathogenic mechanism of periodontitis by leveraging systems biology methods and thoroughly developed a therapeutic option for periodontitis treatment via the prediction of a DNN-based DTI model and drug design specifications.