Drosophila genotypes can be predicted from their exploration locomotive trajectories using supervised machine...

Area of Science:

• Ethology
• Computational Biology
• Genetics

Background:

• Traditional statistical methods may miss insights into fruit fly exploratory activity due to nonlinear locomotive trajectories.
• Understanding genotype-specific behaviors is crucial for fruit fly research.

Purpose of the Study:

• To investigate supervised machine learning's ability to predict fruit fly genotype based on locomotive features.
• To identify key locomotive features that predict genotype during exploratory activity.

Main Methods:

• Captured 10-minute locomotive trajectories of four fruit fly genotypes in an open-field arena.
• Extracted turn angle and step size features from trajectories.
• Trained supervised learning models to predict genotype using these features.

Main Results:

• Achieved 83% accuracy in differentiating wild-type from mutant genotypes using the first five minutes of trajectories.
• Performance decreased when using later or entire trajectory data, indicating early activity is most discriminative.
• Feature importance analysis showed turn angle is a more effective predictor than step size.

Conclusions:

• Locomotive trajectory features can reliably predict fruit fly genotype using supervised machine learning.
• Early exploratory activity (first five minutes) contains the most genotype-discriminative information.
• Machine learning offers a powerful alternative to traditional statistics for analyzing complex behavioral data in fruit flies.