Proteomic Divergence in the Trisomic Mouse Cortex: Machine Learning Identifies Tau, APP, and ADARB1 as Key Genotype Signatures and Reveals Limited Proteomic Response to Memantine

Calvin H. Cho

Keywords: Down syndrome (DS), Alzheimer's disease (AD), Proteomics, Tau-APP Axis, Ts65Dn Mouse Model, Genotype Classification, Memantine, Mouse-Level Split, Replicate Leakage, Elastic Net, Bayesian Logistic Regression, Ensemble Learning, Gradient Boosting, DYRK1A, ITSN1, ADARB1, Collinearity, Signal Integrity.

TL;DR: Using a leakage-free mouse-level evaluation, we identify the Tau-APP axis as the dominant cortical signature of trisomy-AD convergence and find that memantine treatment produces no substantial proteomic reorganization.

Abstract: Down syndrome (DS), caused by trisomy of chromosome 21, confers a near-universal risk of Alzheimer’s disease (AD) pathology by the fifth decade of life. We apply a comprehensive analytical battery to the Mice Protein Expression dataset (77 proteins, 552 samples, Ts65Dn mouse model), including elastic net logistic regression, Bayesian logistic regression with MCMC posterior inference, random forest, and gradient boosting, using a mouse-level train/test split that eliminates technical-replicate leakage. Principal component 1 explains 29.9% of total variance and near-completely separates genotypes without supervision. APP and ITSN1 are the strongest predictors under elastic net penalisation, consistent with chromosome 21 gene dosage; Tau and APP also emerge as dominant full-proteome classifiers across independent ensemble methods (99–100% held-out accuracy). A secondary analysis of memantine treatment within trisomic animals reveals no substantial proteomic reorganisation under the leakage-free evaluation, suggesting memantine’s documented behavioural benefits are pharmacological rather than proteomic in nature. These findings provide rigorous, multi-method characterisation of the molecular convergence between DS and AD-type neurodegeneration, and illustrate how ensemble ML methods expose biological signal obscured by collinearity in classical regression.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 19