DeepTE

DeepTE classifies transposable elements in eukaryotic genomes using convolutional neural networks to assign TE orders and superfamilies for genome annotation and evolutionary analysis.

Key Features:

• Convolutional Neural Networks: DeepTE employs convolutional neural networks (CNNs) to classify transposable element sequences.
• k-mer input encoding: Sequences are transformed into input vectors based on k-mer counts to capture sequence patterns for classification.
• Tree-structured hierarchical classification: Eight distinct models are trained in a tree-structured hierarchy to categorize TEs into superfamilies and orders.
• Domain detection: Domain detection within TEs is incorporated to mitigate false classifications.
• Plant non-TE versus TE model: An additional model is trained specifically to distinguish non-transposable elements from TEs in plant genomes.
• Taxonomic and class coverage: DeepTE classifies unclassified TEs into seven orders and covers 15 superfamilies in plants, 24 in metazoans, and 16 in fungi.

Scientific Applications:

• TE annotation in newly sequenced genomes: Annotating transposable elements in newly sequenced eukaryotic genomes to support genome annotation.
• Evolutionary analyses: Investigating the roles of transposable elements in genome evolution, structure, and function.
• Method benchmarking: Benchmarking and comparative evaluation of TE classification methods.
• Cross-kingdom TE comparison: Comparative analyses of transposable element composition across plants, metazoans, and fungi.

Methodology:

Sequences are converted into k-mer count input vectors; eight CNN models are trained in a tree-structured hierarchy to assign orders and superfamilies; domain detection is used to reduce false classifications; and an additional model distinguishes non-TEs from TEs in plant genomes.