Integrative coexpression and deep learning reveal stage-specific regulators of Apis cerana drone embryogenesis

Abstract

Stage-specific regulators of Apis cerana drone embryogenesis remain poorly defined. RNA-seq profiles from five embryonic stages (12–60 h) were analyzed using weighted gene coexpression network analysis (WGCNA) to construct coexpression modules. Temporal expression patterns were further refined by KNN clustering, and 19 machine learning and deep learning models were benchmarked for gene pattern classification. Top-performing models were integrated through a model consensus intersection strategy and combined with temporal features to identify high-confidence stage-associated candidate genes. Gene expression dynamics were organized into six temporal clusters corresponding to germline and epigenetic regulation (12 h), ncRNA- and rRNA-associated regulation (24 h), neurogenesis and morphogenesis (36–48 h), and late organ specialization (60 h). Among all tested models, ResNet achieved the highest held-out test accuracy. Consensus analysis yielded a robust set of stage-associated genes, from which network central candidate genes and functionally coherent subnetworks were identified. This integrated WGCNA–KNN–ML/DL framework effectively resolves dynamic transcriptional programs during A. cerana drone embryogenesis, revealing coordinated regulatory transitions from early germline specification to late organ development.