Feasibility of predicting allele specific expression from DNA sequencing using machine learning.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Allele specific expression (ASE) describes differing expression levels of alternative alleles, impacting hereditary disease severity.
• Current genome diagnostics rely on DNA sequencing, overlooking crucial gene expression information like ASE.
• Predicting ASE from DNA variation is essential for integrating expression data into diagnostics.

Purpose of the Study:

• To develop and validate machine learning models that predict allele specific expression (ASE) using only DNA variation data.
• To assess the utility of predicted ASE in identifying clinically relevant variants in genes like BRCA2, RET, and NF1.
• To explore the potential of ASE prediction as a novel genome diagnostic tool.

Main Methods:

• Constructed machine learning models using DNA features from BIOS (n=3432) and GTEx (n=369) datasets to predict ASE.
• Applied the trained model to population variants in BRCA2, RET, and NF1 genes.
• Evaluated the model's performance in predicting ASE effects for known pathogenic variants.

Main Results:

• Developed reproducible ASE prediction models incorporating diverse DNA features, indicating complex regulatory mechanisms.
• Successfully predicted ASE effects for 27 variants in key genes, including 10 known pathogenic variants.
• Demonstrated the feasibility of predicting ASE from DNA features using machine learning.

Conclusions:

• Machine learning models can accurately predict allele specific expression (ASE) from DNA variation.
• Predicted ASE holds promise for enhancing genome diagnostics by prioritizing variants for RNA sequencing validation.
• This approach could lead to a new generation of diagnostic tools for hereditary diseases.