Harnessing Deep Learning to Analyze Cryptic Morphological Variability of Marchantia polymorpha.

Area of Science:

• Plant biology
• Genetics
• Computational biology

Background:

• Phenotypic characterization is crucial but challenging in model organisms like Marchantia polymorpha due to morphological variability.
• Distinguishing sexes in M. polymorpha early in development is difficult using traditional methods.

Purpose of the Study:

• To develop and validate a deep-learning-based image classification system for objective sex determination in M. polymorpha.
• To investigate the genetic basis of early sex-linked morphological differences in M. polymorpha.

Main Methods:

• Utilized a deep-learning image classifier trained on M. polymorpha images.
• Established male and female recombinant inbred lines (RILs) to separate sex chromosome effects from autosomal variations.
• Employed an eXplainable AI technique for model validation.

Main Results:

• The deep learning models accurately classified sexes of wild-type M. polymorpha within the first week of growth.
• RILs showed less distinct sex-based morphological differences, suggesting autosomal influence in wild types.
• eXplainable AI confirmed the image regions critical for classification.

Conclusions:

• Deep learning offers a reproducible and objective method for phenotyping M. polymorpha, overcoming limitations of manual feature extraction.
• Early morphological differences in wild M. polymorpha are significantly influenced by autosomal variations.
• This approach is valuable for plant species lacking standardized phenotyping metrics.