Deep learning reveals ecological decoupling and rate heterogeneity in Hymenoptera wing evolution.

HYMENOPTERA WING MORPHOLOGICAL EVOLUTION SHOWS DECOUPLED RATES BETWEEN SHAPE AND VENATION MODULES, UNRELATED TO ECOLOGICAL TRANSITIONS, REVEALED BY DEEP LEARNING.

• Why it matters: Understanding morphological evolution is limited by current methods' scalability and interpretability, hindering insights into ecological and evolutionary processes across diverse taxa.
• What they did: Researchers developed HAIKA, a self-supervised autoencoder that extracts continuous morphological features from insect wing images, and applied it to Hymenoptera to analyze wing shape and venation evolution.
• The result: The study found that wing shape and venation in Hymenoptera evolve at decoupled rates, not consistently linked to ecological shifts like parasitoidism or eusociality, enabling scalable phenotypic analysis.