Deep learning and attention mechanisms to identify key genes and their implications for the origin of insect wings

Abstract

Abstract Wings are a key trait innovation in the evolutionary history of insects, and contributes to the largest diversity of animals on the planet. However, we still have an incomplete understanding of the functional changes in genes behind this diversification. Insect, Malacostraca and Chelicerata species originated as primitive arthropods during the Cambrian period. Malacostraca as the ancestral taxa of winged insects, are key to understanding this radiation. Here, a deep learning (DL) model for wing genes identification (DeepWG) based on bidirectional long short-term memory (BiLSTM) and attention mechanism (AM) was constructed based on the protein sequences of 119 species. DeepWG demonstrated a strong potential in mining key genes of insect wings, achieving an accuracy rate of 97.3% on the test set. Our research found that the 351 key genes identified by DeepWG and their orthologs exhibit transcriptional similarity in wing and gill tissues, providing molecular evidence consistent with the Ttracheal gill theory of insect wing origin. This study not only proposes a new method for identifying key genes, but also lays the foundation for genetic studies of key evolutionary adaptations in winged insects.