Discovering cryptic splice mutations in cancers via a deep neural network framework

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Somatic mutations can disrupt splicing regulatory elements, impacting cancer genes.
• Predicting functional consequences of mutations in extended splice regions is challenging.

Purpose of the Study:

• To characterize the landscape of splice-altering mutations in cancer using deep learning.
• To identify cryptic splice mutations and assess their impact on cancer driver genes.

Main Methods:

• Utilized a deep neural network, SpliceAI, to analyze splice-altering mutations.
• Applied SpliceAI to an in-house cohort of 401 liver cancers and a pan-cancer cohort of 17,714 tumors.
• Validated mutations using matched RNA-sequencing data and performed mutational signature analysis.

Main Results:

• Identified 1244 cryptic splice mutations in liver cancer and over 100,000 in the pan-cancer cohort.
• Uncovered 126 new candidate driver genes and new driver mutations in known cancer genes.
• Revealed increased frequency of splice alterations in tumor suppressor genes and identified mutational processes linked to splice mutations.

Conclusions:

• Deep learning approaches like SpliceAI can effectively identify and annotate functional consequences of cryptic splice mutations in cancer.
• This work enhances driver gene discovery and provides insights into the causes and impact of splice mutations in oncology.