AlzScPred

AlzScPred predicts Alzheimer’s disease status from single-cell transcriptomics data to identify disease-associated genes and support biomarker discovery.

Key Features:

• Data Utilization: Uses gene expression profiles of 33,538 genes across 169,469 cells from Alzheimer’s patients (90,713 cells) and normal controls (78,783 cells).
• Predictive Modeling: Identifies and ranks genes expressed in most cells by their ability to differentiate Alzheimer’s patients from normal controls for feature selection.
• Model Development: Builds machine learning models using selected gene sets of 35 and 100 genes and reports limited generalization of these models on validation datasets due to overoptimization.
• Deep Learning Approach: Applies a deep learning method with dropout regularization to improve robustness, achieving AUCs of 0.75 and 0.84 on validation datasets for the 35- and 100-gene sets, respectively.
• Biological Insights: Performs gene ontology enrichment analysis on selected genes to elucidate molecular mechanisms associated with Alzheimer’s disease.

Scientific Applications:

• AD prediction prototype: Provides a proof-of-concept method for predicting Alzheimer’s disease from single-cell transcriptomics data.
• Biomarker discovery: Supports identification of candidate biomarkers and therapeutic targets for Alzheimer’s disease.
• Early diagnosis and stratification: Aims to inform early diagnosis and personalized treatment strategies through cell-level gene expression signatures.

Methodology:

Initial processing to identify genes expressed across most cells; ranking genes by predictive power for distinguishing Alzheimer’s from controls; training machine learning models on selected 35- and 100-gene sets and applying dropout regularization in a deep learning framework to improve validation performance.