인공지능을 활용한 단백질 구조 및 리간드 결합 모형 분석과 인간 세포 성장 조절에의 적용

초록

This study applied artificial intelligence (AI) tools to analyze protein structures and ligand binding model, and to explore their potential role in regulating human cell growth. Initially, the 3D structure of ATP-binding CDK protein proposed by generative AIs was analyzed with Neurosnap’s Alphafold2 and PyMol protein AIs. Subsequently, 10 hydrophobic ATP analogs were selected as potential CDK inhibitors, and their predicted binding models were examined using Neurosnap’s DiffDock-L AI. These analogs exhibited varying binding affinities, depending on their conformational compatibility and non-covalent interactions with the CDK protein The binding affinities proposed by DiffDock-L were DiffDock Confidence, SMINA Affinity, and SMINA Intramolecular Energy, respectively. Based on these results, we hypothesized that Dinaciclib ATP analogs with higher binding affinities and Flavopiridol analogs with lower affinities would exhibit differential inhibitory effects on CDK proteins, and the effects were further validated using cytological assays in the AGS, HCT-116, and HDF cell lines. Following the treatment of Dinaciclib and Flavopiridol, the cell growth was highly inhibited in a concentration-dependent manner. However, Dinaciclib exhibited significantly (p<0.05) lower half-maximal inhibitory concentration (IC₅₀) values compared to Flavopiridol, indicating higher cytotoxic potency. The cytological results confirmed that Dinaciclib analogs demonstrated more growth inhibition compared to Flavopiridol analogs. These findings suggest that theoretical predictions generated by deep learning-based AI tools are consistent with experimental cytological outcomes, highlighting their powerful potential in life sciences.

키워드