Abstract 5413: AIVE, an artificial intelligence protein miner based on omics data to discover new cancer druggable candidates

Abstract Recently, AI software including Alphafold, RoseTTAFold and ESMFold, can predict protein structures based on amino acid sequences, and DeepMind (Alphafold2) and Meta (ESMFold) opened hundreds of millions of protein structures to world-wide clinical researchers. These tools are possible to generate protein structures in a shorter time and less cost than traditional protein study methods. Finally, they can lead to the advent of the protein rush generation. To evaluate accurately and quickly the prediction of structures of mutations embedding in protein sequences, we developed an Artificial Intelligence analytics toolkit for predicting Virus mutation in protEin (AIVE; http://ai-ve.org) that clinical researchers can use easily on web. An amino acid property eigen selection score (APESS) defined in AIVE is a comprehensive mathematical analysis model, andit was calculated basically by below: 1 artificial intelligence base clustering (AIBC) using predicted aligned error (PAE) and per-residue confidence (pLDDT) values, 2 base level change rate (BLCR)using the probability of occurrence of mutation in each amino acid sequence or nucleotide sequence, 3chemical property eigen score (CPES) using amino acid properties (PH, Residue and Hydrophobic),and 4polarity based feature selection (PBFS) using continuous position of polar amino acids including S, C, N, Q, T, and Y [APESS = AIBC ´ BLCR ´ CPES ´ PBFS]. AIVE also calculated the docking score using AlphaFold to estimate protein to protein interaction (PPI) among hosts (or mutations) and provided a graph as final report showing APESS distribution among mutation group, and evaluation of APESS and docking score. For example, according to APESS score in lung cancer, we can find out the feasibility of structure transformation of H-bond in EGFR protein (L858R) tyrosine kinase and also show the differences of tyrosine kinase affinity between each lung cancer patients. However, for the T790M mutation of EGFR protein, there was no significant difference compared with the wild type (790). The execution time of AIVE (GPU based web server platform) depends on the length of the input sequence, and it completes such an example case in an average of 2-3 hours. AIVE can suggest anti-cancer candidates in protein-level and present a precision medicine model which is available to implement close to real time. Citation Format: Dongwan Hong. AIVE, an artificial intelligence protein miner based on omics data to discover new cancer druggable candidates. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5413.