Impact of Feature Choice on Machine Learning Classification of Fractional Anomalous Diffusion.

Area of Science:

• Biophysics
• Computational Biology
• Machine Learning

Background:

• Machine learning (ML) is increasingly applied to analyze complex biological data.
• Single-particle tracking (SPT) generates large datasets requiring robust analytical methods.
• Classifying anomalous diffusion is essential for understanding molecular dynamics in biological systems.

Purpose of the Study:

• To investigate the impact of feature selection on classifying fractional anomalous diffusion trajectories.
• To compare the performance of different feature sets in random forest and gradient boosting algorithms.
• To evaluate the influence of synthetic training data on classification accuracy.

Main Methods:

• Utilized random forest and gradient boosting machine learning algorithms.
• Compared two existing sets of human-engineered features with a newly developed, problem-specific feature set.
• Assessed the effect of synthetic data variations on classifier performance.
• Validated classifiers using real single-particle tracking data of G proteins and receptors.

Main Results:

• The choice of features significantly impacts the classification accuracy of anomalous diffusion trajectories.
• A tailor-made feature set demonstrated superior performance compared to previously used attributes.
• Alterations in synthetic training data influenced the classification outcomes.
• Classifiers performed effectively on real-world plasma membrane protein trajectory data.

Conclusions:

• Thoughtful feature engineering and selection are critical for successful application of ML in analyzing single-particle tracking data.
• Optimized feature sets enhance the reliability of ML-based trajectory classification.
• The findings provide guidelines for applying ML to biophysical trajectory analysis.