Predicting sites of epitranscriptome modifications using unsupervised representation learning based on generative...

Area of Science:

• * Computational Biology
• * Bioinformatics
• * Molecular Biology

Background:

• * The epitranscriptome, encompassing chemical modifications on RNA transcripts, is a rapidly growing field of study.
• * Predicting the locations of these modifications from RNA sequences is crucial but hindered by a scarcity of positive examples for most modification types.
• * This data imbalance poses significant challenges for developing robust predictive algorithms.

Purpose of the Study:

• * To develop a novel computational approach, MR-GAN, for unsupervised learning of transcriptomic sequence embeddings.
• * To address the challenge of limited positive samples in predicting multiple epitranscriptome modification sites.
• * To improve the accuracy and efficiency of predicting various RNA modifications.

Main Methods:

• * Proposed MR-GAN, a generative adversarial network (GAN) trained in an unsupervised manner on pre-mRNA sequences.
• * Utilized MR-GAN to extract low-dimensional embeddings from transcriptomic sequences for eight epitranscriptome modifications.
• * Trained prediction models using these learned embeddings and compared performance against one-hot encoding and SRAMP.

Main Results:

• * MR-GAN embeddings significantly outperformed one-hot encoding, achieving up to a 15% improvement in prediction accuracy.
• * The model successfully identified known and novel sequence motifs associated with different epitranscriptome modifications.
• * Demonstrated the effectiveness of unsupervised learning for extracting transcriptome features for precise modification site prediction.

Conclusions:

• * Unsupervised learning using MR-GAN provides a powerful strategy for predicting epitranscriptome modifications, especially with imbalanced and limited datasets.
• * The learned embeddings capture essential sequence features, enabling high-precision prediction across multiple modification types.
• * MR-GAN offers a valuable tool for advancing epitranscriptome research and understanding RNA regulation.