lncRNA-MFDL: identification of human long non-coding RNAs by fusing multiple features and using deep learning

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Long noncoding RNAs (lncRNAs) are critical gene regulators involved in cellular functions and disease development.
• High-throughput sequencing has revealed a vast number of novel transcripts, necessitating efficient computational methods for lncRNA identification.
• Distinguishing lncRNAs from coding RNAs is crucial for understanding their biological roles.

Purpose of the Study:

• To develop a powerful and accurate computational predictor for identifying long noncoding RNAs (lncRNAs).
• To enhance the accuracy and robustness of lncRNA prediction compared to existing tools.
• To provide a freely available tool for the scientific community to facilitate lncRNA research.

Main Methods:

• Developed the lncRNA-MFDL predictor integrating features from open reading frame, k-mer, secondary structure, and coding domain sequences.
• Employed deep learning classification algorithms for lncRNA identification.
• Validated the predictor using 10-fold cross-validation on human training datasets and tested on datasets from multiple species.

Main Results:

• lncRNA-MFDL achieved 97.1% prediction accuracy on human datasets, outperforming CPC, CNCI, and lncRNA-FMFSVM by 5.7%, 3.7%, and 3.4%, respectively.
• The predictor demonstrated high effectiveness and robustness across diverse species, including zebrafish, mouse, and C. elegans.
• lncRNA-MFDL significantly improved the accuracy of distinguishing lncRNAs from coding RNAs.

Conclusions:

• lncRNA-MFDL is a highly accurate and robust tool for identifying long noncoding RNAs.
• The developed method offers a significant advancement in computational lncRNA discovery.
• The lncRNA-MFDL software package is available for academic use, supporting further research in the field.