CREATE: a novel attention-based framework for efficient classification of transposable elements.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transposable elements (TEs) are mobile DNA sequences vital for genome regulation and evolution.
• Accurate classification of TEs is essential for understanding their genomic impact.
• Existing alignment-based and machine learning methods have limitations in capturing TE features.

Purpose of the Study:

• To develop a novel framework, CREATE, for efficient and accurate transposable element (TE) classification.
• To address the limitations of existing methods in capturing multiscale features of TEs.
• To improve the accuracy of TE annotation in eukaryotic genomes.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with an Attention mechanism.
• Integrated global pattern distribution and local sequence profiles of TEs.
• Implemented a hierarchical classification strategy with nine parent-node classifiers.

Main Results:

• CREATE demonstrated superior performance compared to existing TE-type annotation methods.
• The framework achieved high accuracy in hierarchical classification tasks.
• CREATE effectively captures both global and local features for improved TE classification.

Conclusions:

• CREATE offers a significant advancement in TE annotation accuracy.
• The proposed framework shows great potential for genomic research and evolution studies.
• CREATE provides an efficient and accurate solution for classifying transposable elements.