Basic Science and Pathogenesis.

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• The AD-BXD mouse panel offers a genetically diverse model for Alzheimer's disease (AD) research, incorporating human familial AD transgenes.
• Cognitive functions, such as contextual fear acquisition (CFA) and contextual fear memory (CFM), are key phenotypes studied.
• Challenges include sparse non-cognitive data requiring imputation and high-dimensional genetic data (SNP) with limited sample sizes.

Purpose of the Study:

• To develop a predictive model for cognitive function in AD-BXD mice.
• To identify potential biomarkers for AD through feature importance analysis.
• To overcome data challenges like missing values and high dimensionality.

Main Methods:

• Utilized bootstrapped k-NN imputation to handle missing non-cognitive phenotype data.
• Employed an autoencoder for single-nucleotide polymorphism (SNP) data dimensionality reduction.
• Trained a multi-layer perceptron (MLP) model iteratively to predict cognitive phenotypes using combined features.

Main Results:

• Achieved a Pearson Correlation Coefficient (PCC) of 0.587 for CFM and 0.504 for CFA using all features.
• Demonstrated improved predictive performance with SNP data alone, yielding PCC of 0.69 for CFM and 0.606 for CFA.
• Identified biomarkers through feature importance analysis consistent with existing literature.

Conclusions:

• The developed model effectively predicts cognitive function in the AD-BXD mouse population.
• Data imputation and SNP dimensionality reduction significantly improved model performance.
• The findings highlight the potential for identifying novel AD biomarkers using this integrated approach.