Computational modeling of mRNA degradation dynamics using deep neural networks

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Messenger RNA (mRNA) degradation is crucial for post-transcriptional gene regulation.
• The 3 acronym{'-UTR elements significantly influence mRNA degradation dynamics.
• Previous computational models relied on a linear degradation assumption, limiting mechanistic understanding.

Purpose of the Study:

• To develop deep neural networks for predicting mRNA degradation dynamics.
• To interpret these networks to identify regulatory elements within the 3 acronym{'-UTR and their positional effects.
• To improve upon existing computational methods for modeling mRNA degradation.

Main Methods:

• Developed deep neural networks (CNNs and RNNs) to predict mRNA levels over time from 3 acronym{'-UTR sequences.
• Utilized Integrated Gradients for model interpretability to identify cis-regulatory elements.
• Employed mutagenesis analysis to investigate the positional effects of 3 acronym{'-UTR elements.
• Compared single-task and multi-task learning models, including those with different poly(A) tail lengths.

Main Results:

• Deep neural networks significantly outperformed existing methods in predicting mRNA degradation dynamics.
• Convolutional neural network (CNN) models demonstrated superior interpretability compared to recurrent neural networks (RNNs).
• Identified known and novel cis-regulatory sequence elements and their positional effects on mRNA degradation.
• Multi-task learning models, considering poly(A) tail variations, showed improved performance.

Conclusions:

• Deep neural networks provide a powerful tool for understanding mRNA degradation dynamics.
• Identified specific 3 acronym{'-UTR elements and their positions critical for gene regulation.
• The study enhances mechanistic insights into mRNA decay pathways.