Classification of particle trajectories in living cells: Machine learning versus statistical testing hypothesis for...

Area of Science:

• Cell biology
• Biophysics
• Computational biology

Background:

• Single-particle tracking (SPT) is crucial for studying intracellular transport dynamics.
• Understanding these dynamics is key to cell organization and function.
• Current methods like mean-square displacement (MSD) have limitations in trajectory classification.

Purpose of the Study:

• To apply machine learning (ML) ensemble algorithms for classifying single-particle tracking data.
• To introduce novel features for transforming raw trajectory data into ML-compatible input vectors.
• To compare ML-based classification performance against existing statistical methods.

Main Methods:

• Utilized two ML ensemble algorithms: random forest and gradient boosting.
• Developed a new feature set for trajectory data representation.
• Applied the trained ML models to real-world SPT data from G protein-coupled receptors and G proteins.

Main Results:

• The developed ML models demonstrated robust classification of particle trajectory types.
• The novel feature set enabled effective transformation of raw SPT data.
• Performance was evaluated against advanced statistical methods beyond MSD.

Conclusions:

• Machine learning, particularly random forest and gradient boosting, provides a powerful framework for trajectory classification in SPT data.
• The proposed feature engineering approach enhances the applicability of ML to biological transport analysis.
• This study offers a more advanced alternative to traditional MSD-based classification for understanding cellular processes.