Deep multiple-instance learning accurately predicts gene haploinsufficiency and deletion pathogenicity.

Area of Science:

• Genetics and genomics
• Computational biology
• Bioinformatics

Background:

• Copy number losses (deletions) are significant causes of severe genetic disorders.
• Current prediction methods struggle to integrate multiple lines of evidence for gene haploinsufficiency, limiting their ability to identify pathogenic deletions.

Approach:

• Introduced DosaCNV, a deep multiple-instance learning framework.
• Models deletion pathogenicity jointly with gene haploinsufficiency.
• Integrates over 30 gene-level features predictive of haploinsufficiency.

Key Points:

• DosaCNV demonstrates superior performance in prioritizing pathogenic deletions across various genetic disorders.
• Outperforms existing methods in predicting gene haploinsufficiency without prior training on known haploinsufficient genes.
• Utilizes advanced techniques for quantifying feature contributions to haploinsufficiency, enabling interpretable predictions.

Conclusions:

• DosaCNV is a powerful computational tool for genetic disorder research.
• Enhances both fundamental understanding and translational applications in identifying disease-causing deletions.