DEMINING: A deep learning model embedded framework to distinguish RNA editing from DNA mutations in RNA sequencing...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Distinguishing RNA editing from DNA mutations is crucial for transcriptomic analysis but challenging due to errors.
• Adenosine-to-inosine (A-to-I) RNA editing is a common post-transcriptional modification.
• DNA mutations can be mistaken for RNA editing events in sequencing data.

Purpose of the Study:

• To develop a computational framework (DEMINING) to accurately differentiate RNA editing sites from DNA mutations.
• To apply the framework to identify novel RNA editing and DNA mutation sites in patient samples.
• To investigate the functional implications of identified sites in acute myeloid leukemia.

Main Methods:

• Developed a stepwise computational framework named DEMINING.
• Integrated a deep learning model (DeepDDR) for classification.
• Utilized transfer learning for application to non-primate samples.
• Applied the framework to RNA sequencing data from acute myeloid leukemia patients.

Main Results:

• DEMINING successfully distinguishes RNA editing from DNA mutations directly from RNA sequencing data.
• The tool identified previously underappreciated DNA mutation and RNA editing sites in leukemia.
• Some identified sites were associated with upregulated host gene expression.
• Several sites were linked to the production of neoantigens.

Conclusions:

• DEMINING provides a robust method for accurate identification of RNA editing and DNA mutations.
• The framework enhances the analysis of transcriptomic data by resolving confounding mutations.
• The findings in acute myeloid leukemia highlight the clinical relevance of precise RNA editing and mutation profiling.