Using expression quantitative trait loci data and graph-embedded neural networks to uncover genotype-phenotype...

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Establishing a functional genotype-phenotype map is a central goal in biology.
• Multi-omics analysis offers opportunities to understand genotype-phenotype correlations but faces challenges like insufficient omics types, unclear internal correlations, and small sample sizes relative to traits (n << p).
• These challenges hinder the application of machine learning for disease outcome classification.

Purpose of the Study:

• To address limitations in multi-omics analysis for genotype-phenotype mapping.
• To propose a robust classification model for disease outcomes.
• To develop a method that accounts for internal omics correlations and prevents overfitting.

Main Methods:

• A graph-embedded deep neural network (G-EDNN) was proposed, utilizing expression quantitative trait loci (eQTL) data.
• The G-EDNN incorporates sparse connectivity between network layers to mitigate overfitting.
• Internal correlations within each omics type were considered to enhance biological realism.
• Experimental analysis was performed using GSE28127 and GSE95496 datasets from the Gene Expression Omnibus (GEO) database.
• Feature selection and graph embedding were performed using prior data.

Main Results:

• The proposed G-EDNN method demonstrated high classification accuracy.
• The method facilitates easy-to-interpret feature selection.
• The approach effectively addresses challenges in multi-omics data analysis.

Conclusions:

• The G-EDNN provides a robust framework for genotype-phenotype association analysis.
• This method extends the application of deep learning in understanding genotype-phenotype relationships.
• The approach offers improved disease classification and feature selection in multi-omics studies.