演題概要

オーラルセッション

第3日　6月24日（金）　10:00～10:20　A会場（メインホール）

3A-O1-1000　　PDF

多層オミックスと機械学習による軽度認知障害の予測モデルの構築

(¹弘大院医・ ²東北大メガバンク・ ³弘大院保健・ ⁴HMT)
^o多田羅洋太¹・ 山嵜博未¹・ 勝岡史城²・ 千葉満³・ 三枝大輔²・ 葛西秋宅¹・ 中村智洋²・ 亀谷直孝⁴・ 山本博之⁴・ 氏原大⁴・ 藤本哲太⁴・ 荘司美穂⁴・ 元池育子²・ 玉田嘉紀¹・ 橋爪克仁⁴・ 東海林幹夫¹・ 木下賢吾²・ 中路重之¹・ 山本雅之²・ 伊東健¹

Since dementia is preventable with early interventions, biomarkers that assist in diagnosing early stages of dementia, such as mild cognitive impairment, are urgently needed. Multiomics analysis of amnestic MCI (aMCI) peripheral blood (n=25) was performed covering the transcriptome, miRNA, proteome, and metabolome. Validation analysis for miRNAs was conducted in an independent cohort (n=12). Artificial intelligence was used to identify the most important features for predicting aMCI. We found that hsa-miR-4455 is the best biomarker in all omics analyses. The diagnostic index taking a ratio of hsa-miR-4455 to hsa-let-7b-3p predicted aMCI patients against healthy subjects with 97% overall accuracy. An integrated review of multiomics data suggested that a subset of T cells and amino acid starvation stress response are associated with aMCI. This study proposes a framework for generating new hypotheses including additional research, i.e., large-scale studies to validate biomarkers for clinical use and to clarify functions of the miRNAs.