Aberrant gene expression prediction across human tissues

Area of Science:

• Genomics
• Computational Biology
• Precision Medicine

Background:

• Aberrant gene expression is frequently implicated in diseases.
• Current algorithms for predicting individual aberrant gene expression are lacking.

Purpose of the Study:

• To develop and validate a predictive model for aberrant gene expression.
• To improve gene discovery and phenotype prediction using variant data.

Main Methods:

• Compiled a benchmark of 8.2 million rare variants from 633 individuals across 49 tissues.
• Trained the AbExp model integrating variant annotations, expression variability, and splicing effects.
• Integrated expression data from clinically accessible tissues.

Main Results:

• Existing tools (CADD, LOFTEE) showed mild predictive ability (1-1.6% average precision).
• The AbExp model achieved 12% average precision, with a two-fold improvement when integrating accessible tissue data.
• AbExp variant scores increased gene discovery sensitivity and improved phenotype predictions on UK Biobank blood traits.

Conclusions:

• AbExp offers a significant advancement in predicting aberrant gene expression.
• The model's tissue-specific approach and integration of diverse data enhance its utility.
• This method holds promise for advancing genetic disease research and personalized medicine.