SkyNet For Drugs: A Novel User-Centric Approach to Drug Discovery

Traditional drug discovery is an expensive and laborious multi-step process that requires a detailed understanding of disease pathobiology, potential drug target characterization, synthesis, experimental evaluation, and optimization of putative drug candidates as a pretext for clinical evaluation which often does not translate into success. With the advent of whole genome sequencing, machine learning, and artificial intelligence, drug discovery and development are now enhanced both in speed and precision. SkyNet For Drugs (Skynet4D) is an AI-driven platform that automates the preclinical processes by integrating drug-target retrieval based on patient's diagnostic medical reports, and chemical database/ADMETox screening leading to the selection of a potential drug candidate. Skynet4D framework utilizes a combination of Natural Language Processing (NLP-GenAI), custom molecular docking tools (DiffDock®-GNINA®) for high-throughput virtual screening (HTVS), and an AI-enabled ADMETox (ADMETAi®) tool to generate drug candidates with a high chance of being effective within clinical settings. In this paper, the proof-of-concept for Skynet4D was presented with a medical diagnostic report of a COVID-19 patient; leading to the prediction of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS‑CoV‑2) main protease, betaCoV_Nsp5_Mpro as putative targets. The target sequences retrieval, 3D structures retrieval/generation, putative ligand search (COCONUT, LOTUS, ChEMBL, etc.), and ranking were done autonomously within the Skynet4D pipeline. DiffDock docking confidence score of ≥ 1 and a GNINA binding affinity score of ≤ -6.0 kcal/mol, signaled suitable ligands for selection within the DiffDock®-GNINA®. The best-ranked compounds were filtered with ADMETAi® which ultimately gave Lansoprazole as the candidate compound. Keywords: Drug discovery, Computer-aided drug discovery, Structure-based drug design, Skynet4D, High-throughput virtual screening, Patient-centric drug discovery SkyNet For Drugs: A Novel User-Centric Approach to Drug Discovery 1*Obi, E.D., 1Yentumi, J.A., 2Ajayi O.O., 3Omotuyi, O.I. & 1Ashimolowo, S.B. 1 Autogon Inc. 3002 Falls at Fairdale, 77057, USA 2Department of Computer Science, Adekunle Ajasin University, Nigeria 3Department of Pharmacology and Toxicology, College of Pharmacy, Afe Babalola University, Ado-Ekiti, Ekiti State, Nigeria E-mail: [email protected], [email protected], [email protected], [email protected], [email protected]; Phone: 1*+1 832 925 1036, 1+233 54 100 6410, 2+27 64 0499906, 3+234 807 094 3256, 1+234 803 527 8071; ABSTRACT Traditional drug discovery is an expensive and laborious multi-step process that requires a detailed understanding of disease pathobiology, potential drug target characterization, synthesis, experimental evaluation, and optimization of putative drug candidates as a pretext for clinical evaluation which often does not translate into success. With the advent of whole genome sequencing, machine learning, and artificial intelligence, drug discovery and development are now enhanced both in speed and precision. SkyNet For Drugs (Skynet4D) is an AI-driven platform that automates the preclinical processes by integrating drug-target retrieval based on patient's diagnostic medical reports, and chemical database/ADMETox screening leading to the selection of a potential drug candidate. Skynet4D framework utilizes a combination of Natural Language Processing (NLP-GenAI), custom molecular docking tools (DiffDock®-GNINA®) for high-throughput virtual screening (HTVS), and an AI-enabled ADMETox (ADMETAi®) tool to generate drug candidates with a high chance of being effective within clinical settings. In this paper, the proof-of-concept for Skynet4D was presented with a medical diagnostic report of a COVID-19 patient; leading to the prediction of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS‑CoV‑2) main protease, betaCoV_Nsp5_Mpro as putative targets. The target sequences retrieval, 3D structures retrieval/generation, putative ligand search (COCONUT, LOTUS, ChEMBL, etc.), and ranking were done autonomously within the Skynet4D pipeline. DiffDock docking confidence score of ≥ 1 and a GNINA binding affinity score of ≤ -6.0 kcal/mol, signaled suitable ligands for selection within the DiffDock®-GNINA®. The best-ranked compounds were filtered with ADMETAi® which ultimately gave Lansoprazole as the candidate compound. Keywords: Drug discovery, Computer-aided drug discovery, Structure-based drug design, Skynet4D, High-throughput virtual screening, Patient-centric drug discovery