DNABERT-based explainable lncRNA identification in plant genome assemblies.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long non-coding RNAs (lncRNAs) are crucial regulators of plant gene expression, influencing epigenetic and transcript levels.
• Current machine learning methods for lncRNA detection often rely on transcriptome data and manually defined features, leading to underrepresentation and bias.

Purpose of the Study:

• To develop and evaluate Natural Language Processing (NLP) models for identifying plant lncRNAs directly from genomic sequences.
• To assess the accuracy and cross-species prediction capabilities of NLP models for lncRNA discovery.

Main Methods:

• Utilized NLP models trained on genomic sequences from seven plant species: Zea mays, Arabidopsis thaliana, Brassica napus, Brassica oleracea, Brassica rapa, Glycine max, and Oryza sativa.
• Evaluated model performance based on prediction accuracy and explored cross-species prediction potential.
• Applied explainable artificial intelligence to identify sequence motifs critical for lncRNA prediction.

Main Results:

• Achieved high prediction accuracies for lncRNAs from genomic sequences, with the highest at 83.4% for Zea mays and the lowest at 57.9% for Brassica rapa.
• Demonstrated successful cross-species prediction with an average accuracy of 63.1% for unseen species.
• Identified sequence motifs flanking lncRNA regions that are important for prediction, suggesting their functional relevance.

Conclusions:

• NLP models provide an effective and accurate method for identifying plant lncRNAs directly from genomic data, overcoming limitations of transcriptomic approaches.
• Genome assembly quality appears to influence lncRNA identification accuracy.
• NLP models show promise for broad lncRNA discovery across diverse plant species and can be interpreted to reveal key sequence features.