Looking for Mimicry in a Snake Assemblage Using Deep Learning.

Area of Science:

• Evolutionary Biology
• Ecology
• Computer Science

Background:

• Batesian mimicry is a classic example of natural selection, often involving brightly colored species.
• Mimicry in inconspicuous species is less studied, leaving its frequency unknown.
• Previous methods for quantifying visual resemblance are limited.

Purpose of the Study:

• To quantify the frequency of Batesian mimicry in a vertebrate community.
• To investigate mimicry in inconspicuous species.
• To develop a novel deep learning approach for assessing visual resemblance.

Main Methods:

• Developed a deep learning model to quantify visual resemblance from photographs.
• Applied the method to Western Palearctic snakes.
• Identified potential mimics by comparing resemblance to sympatric venomous snakes against random expectations.

Main Results:

• Identified 8% of non-venomous snake species as potential Batesian mimics.
• Mimicry was often subtle, involving imperfect resemblance.
• Concealed species can act as models for mimicry.

Conclusions:

• Deep learning provides a powerful tool for quantitative studies of phenotypic resemblance.
• Batesian mimicry is more widespread than previously thought, even in cryptic species.
• This approach can be applied to detect mimicry in other communities.