Detection and classification of long terminal repeat sequences in plant LTR-retrotransposons and their analysis using...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long terminal repeats (LTRs) are vital components of LTR retrotransposons and retroviruses, abundant in eukaryotic genomes.
• LTRs contain regulatory sequences essential for retrotransposon life cycles, but their structure and composition remain incompletely understood.
• Previous studies offered limited insights, primarily from model systems, necessitating advanced analytical approaches.

Purpose of the Study:

• To enhance understanding of LTR structure and composition by analyzing contrasts between different retrotransposon families and genomic regions.
• To develop and apply machine learning methods for accurate classification and prediction of LTR sequences.
• To identify biologically relevant sequence motifs within LTRs using explainability analysis.

Main Methods:

• Applied machine learning techniques, including Gradient Boosting, hybrid convolutional/long and short memory networks, and transformer-based models, to a large dataset of plant LTR retrotransposon sequences.
• Utilized k-mer sequence representation for DNA pre-trained transformer models.
• Performed explainability analysis on trained models to identify sequence features and their positional relevance.

Main Results:

• All three machine learning approaches successfully classified and isolated LTRs, providing insights into their sequence composition.
• The hybrid network model achieved the highest LTR detection F1 score of 0.85.
• Explainability analysis identified biologically relevant motifs, including a central TATA-box and TG..CA patterns at LTR edges, and highlighted the significance of LTR termini.

Conclusions:

• Machine learning models accurately recognized biologically relevant motifs, such as core promoter elements and transcription factor binding sites.
• Explainability analysis underscored the importance of LTR 5' and 3' edges for identity, suggesting the need for analysis beyond dinucleotides.
• The study demonstrates the utility of machine learning in regulatory sequence analysis and classification, confirming the role of identified motifs in LTR detection.