# Genes, shells, and AI: using computer vision to detect cryptic morphological divergence between genetically distinct populations of limpets

## TL;DR

This paper shows how AI can detect hidden physical differences in limpets that match genetic differences, helping to better understand biodiversity.

## Contribution

The study introduces a scalable AI workflow to uncover cryptic morphological divergence in genetically distinct limpet populations.

## Key findings

- A fine-tuned convolutional network classified limpets into genetic clades with high accuracy (median F1-scores up to 0.96).
- Saliency maps identified specific morphological features like keyholes and ridge tips as clade-specific markers.
- Shape analyses confirmed significant morphological divergence between clades previously thought to be similar.

## Abstract

Many species are composed of two or more genetically distinct clades, indicating ongoing or past evolutionary divergence. Often however, there are no obvious morphological differences between clades, making it difficult to accurately assess specific aspects of biodiversity or to enact targeted conservation efforts. New advancements in artificial intelligence tools can be used to categorise individuals into their respective genetic clades and to highlight their distinguishing morphological characters that would otherwise be hidden from human observers. Here, we applied computer vision and explainable artificial intelligence techniques to four limpet species that display well-defined phylogeographic breaks along the Baja California and California coasts. A fine-tuned convolutional network, trained and evaluated over 100 resampling iterations, classified individuals into their genetic clades with median F1-scores of up to 0.96. F1-score performance was markedly higher for true clade groups than the controlled mixed-groups, confirming the presence of features specific to the clades. Saliency maps consistently emphasised structures such as the keyhole in Fissurella volcano and the ridge tips in Lottia conus as distinguishing features, and subsequent shape analyses confirmed significant divergence between clades. These results demonstrate the power of computer vision and explainable artificial intelligence to expose otherwise cryptic morphological diversity and provide a scalable, reproducible workflow that can broaden the biodiversity toolkit and refine eco-evolutionary research across taxa.

The online version contains supplementary material available at 10.1038/s41598-025-30613-1.