A deep recurrent neural network discovers complex biological rules to decipher RNA protein-coding potential.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The rapid increase in RNA transcript identification necessitates advanced methods for genome annotation.
• Traditional RNA classification relies on predefined features, limiting discovery of novel biological rules.

Purpose of the Study:

• To develop a deep learning model for accurate prediction of RNA protein-coding potential.
• To explore the capability of recurrent neural networks (RNNs) in discovering biological patterns de novo.

Main Methods:

• A gated recurrent neural network (RNN) was trained on human messenger RNA (mRNA) and long noncoding RNA (lncRNA) sequences.
• The model, termed mRNA RNN (mRNN), was evaluated against state-of-the-art methods for predicting coding potential.

Main Results:

• mRNN achieved superior performance in predicting protein-coding potential compared to existing methods.
• The model learned complex, long-range patterns in full-length human transcripts without prior feature knowledge.
• Analysis of the network revealed context-sensitive codons indicative of coding potential.

Conclusions:

• Gated RNNs are effective for complex classification tasks involving large-scale transcriptomic data.
• mRNN offers a powerful tool for enhancing genome annotation and driving biological knowledge discovery.
• This deep learning approach can effectively mine new biological insights from extensive sequencing datasets.