Multimodal zero-shot learning of previously unseen epitranscriptomes from RNA-seq data.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Epitranscriptome analysis is crucial for understanding RNA modification dynamics.
• Current prediction methods require condition-specific epitranscriptome data, limiting their application.
• Epitranscriptome profiling is technically challenging and costly, restricting data availability.

Purpose of the Study:

• To develop a multimodal zero-shot learning framework (ExpressRM) for predicting condition-specific RNA modifications.
• To enable prediction in previously unseen contexts without matched epitranscriptome data.
• To expand the applicability of RNA modification prediction to broader biological scenarios.

Main Methods:

• ExpressRM utilizes genome and RNA-sequencing (RNA-seq) data.
• It employs a multimodal zero-shot learning approach.
• The framework does not require epitranscriptome data for training.

Main Results:

• ExpressRM accurately predicts epitranscriptomes in unseen conditions using only transcriptome data.
• Performance is comparable to existing in-condition learning algorithms.
• The method can distinguish between dynamic and static RNA methylation sites.
• A case study identified N6-methyladenosine sites in glioblastoma, revealing pathological insights.

Conclusions:

• ExpressRM effectively leverages transcriptome data for RNA modification prediction in novel contexts.
• The framework broadens the scope of epitranscriptome research where RNA-seq is available.
• It offers valuable insights into the dynamic roles of RNA modifications in various diseases and biological states.