A self-attention-based deep learning model for identifying key genes in insect pupal metamorphosis

Area of Science:

• Insect biology
• Evolutionary developmental biology
• Bioinformatics

Background:

• Insect metamorphosis is a key evolutionary innovation critical for adaptation and biodiversity.
• Identifying genes that regulate insect pupal metamorphosis remains a significant challenge in biological research.

Purpose of the Study:

• To develop and validate a novel deep learning approach for identifying key genes in insect metamorphosis.
• To elucidate the functional roles of identified key genes during the pupal stage.

Main Methods:

• Construction of a Word2vec-based word embedding library from 293 insect protein sequences across 14 orders.
• Design of a Gene Network Classification Model (GNCM) integrating deep learning (DL) and self-attention mechanisms (SAM).
• Comparative performance analysis of GNCM against traditional machine learning algorithms (ANN, SVM, XGBoost) and recurrent neural networks (BiGRU, BiLSTM).

Main Results:

• GNCM demonstrated superior classification accuracy and interpretability compared to other models.
• Identified 1,048 high-weight gene families crucial for insect pupal metamorphosis.
• High-weight genes showed significantly higher expression during the pupal stage and were associated with development, apoptosis, and immunity pathways (KEGG annotation).

Conclusions:

• The novel GNCM approach effectively identifies key genes regulating insect metamorphosis.
• The identified genes play critical roles in developmental, apoptotic, and immune processes essential for metamorphosis.
• This study offers new insights into the genetic mechanisms underlying insect metamorphosis and provides a powerful AI tool for future research.