勷勤数学•专家报告

题      目：A Comparative Study of Blood-Based Transcriptomic and Metabolomic Signatures Alzheimer's Disease

报  告  人：李岩 博士  (邀请人：张旭)

                                    加州大学欧文分校

时      间： 12月2日  10:00-11:00

地     点：数科院二楼会议室

报告人简介:

       李岩，加州大学欧文分校流行病学与生物统计系博士后，2023年博士毕业于东北师范大学统计系，2023～2025年在普度大学统计系和加州大学流行病学与生物统计系做博士后。目前主要研究方向为生物信息学，包括代谢组学，空间转录学，单细胞数据分析以及相关机器学习和统计学方法研究。在Statistics and Probability Letters， Alzheimer's & Dementia 等杂志发表多篇论文。

摘      要：

        We conducted a comparative study of AD detection using the ANMerge cohort of 192 individuals (98 healthy controls and 94 AD patients), which includes demographic information, transcriptomic and metabolomic profiles obtained through three chromatographic modes (HILIC ESI+, RPC ESI+, RPC ESI-). We trained tree-ensemble models AD prediction and assessed performance using validation AUC. Demographics alone achieved AUC 66.3% but added little when other data were available: transcriptomics (76.0%) vs. demographics + transcriptomics (76.3%); HILIC ESI+ (91.7%) vs. HILIC ESI+ + demographics (90.5%). The best performance was the metabolomics model combining HILIC ESI+ and RPC ESI- (AUC 93.0%). Selecting variables with Q-SHAP algorithm, we achieved AUC over 90% with an average of 22.5±1.44 features in demographics + HILIC ESI+ + RPC ESI- (90.7%).While transcriptomic profiles reflect disease states, the metabolomic profile instantly reacts to any change in the dynamic physiological states and is therefore more sensitive to disease status perturbations.

        欢迎老师、同学们参加、交流！