Predicting Splicing from Primary Sequence with Deep Learning.

Area of Science:

• Molecular Biology
• Genetics
• Bioinformatics

Background:

• Splicing precision is crucial for gene expression, yet the underlying mechanisms remain unclear.
• Cryptic splicing, triggered by genetic variants, can lead to disease but is difficult to predict.
• Understanding splice junction prediction is vital for diagnosing genetic disorders.

Purpose of the Study:

• To develop a deep neural network for accurate splice junction prediction.
• To identify noncoding genetic variants causing cryptic splicing.
• To assess the role of splice-altering variants in human diseases.

Main Methods:

• Developed a deep neural network model for splice junction prediction.
• Analyzed synonymous and intronic mutations for splice-altering consequences.
• Validated predictions using RNA-sequencing data.
• Examined de novo mutations in patients with autism and intellectual disability.

Main Results:

• The deep neural network accurately predicts splice junctions from pre-mRNA sequences.
• Predicted splice-altering mutations show high validation rates with RNA-seq and are deleterious in the human population.
• De novo mutations with splice-altering consequences are enriched in patients with autism and intellectual disability.
• RNA-seq validated predicted splice-altering mutations in 21 out of 28 patients.

Conclusions:

• Deep neural networks can precisely predict splice junctions and identify cryptic splicing.
• Splice-altering genetic variants represent a significant, underappreciated cause of rare genetic disorders and neurodevelopmental conditions.
• This approach aids in diagnosing genetic diseases and understanding mutation impacts.