Prediction of Long Non-Coding RNAs Based on Deep Learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing has generated vast transcript data, necessitating efficient methods for long non-coding RNA (lncRNA) identification.
• Accurate lncRNA detection is crucial for understanding biological processes and molecular-level disease mechanisms.
• Current computational and experimental methods for lncRNA detection face limitations due to sequencing technology constraints and inherent errors.

Purpose of the Study:

• To develop and evaluate a deep learning model for distinguishing long non-coding RNAs (lncRNAs) from messenger RNAs (mRNAs).
• To improve the accuracy and efficiency of lncRNA identification compared to existing methods.

Main Methods:

• Constructed a deep learning framework integrating a bidirectional long short-term memory (BLSTM) layer and a convolutional neural network (CNN) layer.
• Utilized k-mer embedding vectors, trained via the GloVe algorithm, as input features for the model.
• Compared the model's performance against traditional methods like PLEK, CNCI, and CPC.

Main Results:

• The deep learning model achieved high classification performance with an F1-score of 97.9%, accuracy of 96.4%, and auROC of 99.0%.
• Demonstrated superior performance in distinguishing lncRNAs from mRNAs compared to PLEK, CNCI, and CPC methods.
• The model effectively identified lncRNAs, highlighting its potential in molecular diagnostics.

Conclusions:

• The developed deep learning model offers a robust and effective approach for differentiating lncRNAs from mRNAs.
• This model shows promise as a tool for advancing research into lncRNA-associated diseases.
• The findings contribute to the ongoing challenge of accurate lncRNA identification in transcriptomic data.