Classifying Alzheimer’s Disease Outcome in Single-Cell RNA-Seq Datasets

Overview

Alzheimer’s disease (AD), a neurodegenerative disease inflicting nearly 6 million people in the United States, is characterized by molecular and cellular changes in the brain that result in cognitive impairment. Previous studies have applied machine learning methods to predict AD in bulk RNA datasets. However, since AD’s hallmark pathological characteristics are deviations from regular cellular functions, interpreting ML models trained on single cell RNA sequencing (scRNA-seq) datasets could reveal a relationship between age-related features from genomics datasets and Alzheimer’s disease outcome. In this study, we tested whether Alzheimer’s disease can be predicted on the single cell level by building a machine learning model trained on scRNA-seq datasets. Simultaneously, we built on previous laboratory work to test the effects of count binarization on model performance. We present two models that can successfully predict Alzheimer’s Disease for held-out samples from the same patient datasets they have been trained on, with an AUROC score of up to 0.9752. Even more interestingly, these models generalize across patients, and can classify samples as control or diseases from never-seen-before patient datasets. Our subsequent analysis, which involved extracting and ranking principal component loadings, highlighted 100 top genes, such as LY6H, DYNLL1, ERBB4, and NEAT1 – all of which have previously been connected to AD. Further analysis of extracted features could shed light on the relationship between gene expression and AD progression, which may aid in early AD diagnosis.

Competition history

• AJAS 2024 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Biology and Bioinformatics projects
• AJAS Computational Biology and Bioinformatics projects
• Every AJAS 2024 project