Prediction of plant lncRNA by ensemble machine learning classifiers

Area of Science:

• Plant molecular biology
• Bioinformatics
• Genomics

Background:

• Long non-protein coding RNAs (lncRNAs) play crucial roles in plant development and stress responses.
• Research on plant lncRNAs is hindered by a lack of effective prediction tools and validated databases.
• Existing prediction models often rely on non-validated transcripts, primarily focusing on animal systems.

Purpose of the Study:

• To develop and validate a novel computational tool for predicting plant long non-coding RNAs.
• To improve the accuracy and reliability of identifying candidate lncRNAs for functional studies.
• To create a resource that aids in the elucidation of lncRNA functions in plants.

Main Methods:

• Constructed individual stochastic gradient boosting and random forest classifiers using empirically validated lncRNAs from plant, animal, and viral sources.
• Developed an ensemble approach by combining multiple classifiers into a single stacking meta-learner.
• Utilized transcript sequence features for predicting putative plant lncRNAs.

Main Results:

• The ensemble classifier effectively identified putative plant lncRNAs.
• Comparison with the GreeNC database showed significant overlap (51-83%) for predicted genes in *Arabidopsis thaliana*, *Oryza sativa*, and *Eutrema salsugineum*.
• Highest-ranking predictions in *Arabidopsis thaliana* included potential natural antisense genes, pseudogenes, and transposable elements.

Conclusions:

• The developed ensemble classifier is an accurate tool for ranking long non-protein coding RNA predictions.
• This tool can be integrated with gene expression studies to advance research on lncRNA function.
• Accurate identification of plant transcripts with regulatory potential as lncRNAs will facilitate future functional research.