Revealing the grammar of small RNA secretion using interpretable machine learning

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Small non-coding RNAs are secreted via exosomes and lipoproteins.
• Mechanisms governing small RNA sorting and secretion remain largely unknown.

Purpose of the Study:

• To develop a machine learning model for predicting small RNA secretion.
• To identify factors influencing small RNA secretion.
• To explore therapeutic and synthetic biology applications of small RNA secretion.

Main Methods:

• Developed ExoGRU, a machine learning model for predicting small RNA secretion probabilities from RNA sequences.
• Performed experimental validation using ExoGRU-guided mutagenesis and synthetic RNA analysis.
• Developed exoCLIP technique to analyze RNA interactome of RNA-binding proteins in cell-free space.

Main Results:

• ExoGRU accurately predicts small RNA secretion.
• Identified cis and trans factors, including RNA-binding proteins (RBPs) like YBX1, HNRNPA2B1, and RBM24, involved in small RNA secretion.
• Discovered novel biological mechanisms governing small RNA secretion.

Conclusions:

• Machine learning, exemplified by ExoGRU, is powerful for uncovering biological mechanisms.
• Findings provide insights into small RNA secretion for potential therapeutic and synthetic biology uses.