[Paper Review] Machine learning method for single trajectory characterization

Gorka Muñoz-Gil, Miguel Ángel García-March+3 more|arXiv (Cornell University)|Mar 7, 2019

Diffusion and Search Dynamics37 references41 citations

TL;DR

A Random Forest-based approach classifies single-particle trajectories by diffusion model and estimates the anomalous diffusion exponent, robust to short length and noise, with transfer learning to experimental data.

ABSTRACT

In order to study transport in complex environments, it is extremely important to determine the physical mechanism underlying diffusion, and precisely characterize its nature and parameters. Often, this task is strongly impacted by data consisting of trajectories with short length and limited localization precision. In this paper, we propose a machine learning method based on a random forest architecture, which is able to associate even very short trajectories to the underlying diffusion mechanism with a high accuracy. In addition, the method is able to classify the motion according to normal or anomalous diffusion, and determine its anomalous exponent with a small error. The method provides highly accurate outputs even when working with very short trajectories and in the presence of experimental noise. We further demonstrate the application of transfer learning to experimental and simulated data not included in the training/testing dataset. This allows for a full, high-accuracy characterization of experimental trajectories without the need of any prior information.

Motivation & Objective

• Characterize single trajectories to identify underlying diffusion models (CTRW, FBM, LW, ATTM).
• Estimate the anomalous diffusion exponent alpha from single trajectories.
• Demonstrate robustness to short trajectory length and measurement noise.
• Show transfer learning capability from simulated to experimental data.

Proposed method

• Transform trajectories into a standardized preprocessing representation to enable scale-invariant analysis.
• Train a Random Forest on simulated trajectories from CTRW, FBM, Lévy walks, and ATTM to classify diffusion models.
• Use RF regression to predict the anomalous exponent alpha from single trajectories.
• Apply preprocessing that normalizes displacements and constructs a normalized trajectory for RF input.
• Demonstrate robustness to noise and short trajectory length, and perform transfer learning to experimental datasets.

Experimental results

Research questions

• RQ1 Can a Random Forest accurately discriminate among diffusion models from single, short trajectories?
• RQ2 Can the RF reliably estimate the anomalous diffusion exponent alpha from single trajectories, including nonergodic cases?
• RQ3 How robust is the approach to noise and limited trajectory length?
• RQ4 Can the model transfer learn from simulated data to experimental single-trajectory datasets?

Key findings

• RF achieves high accuracy in discriminating diffusion models, especially when preprocessing is used to maintain short-time features.
• The RF can predict the anomalous exponent with MAE around 0.11 for tmax=1000 in noise-free subdiffusive data, with ~80% of predictions within 0.1 of the true value.
• Model discrimination accuracy remains relatively high for short trajectories (10 points) with modest degradation.
• RF predictions remain robust to Gaussian localization noise up to sigma_n close to 1, with increasing error at higher noise.
• Transfer learning successfully classifies experimental datasets (e.g., compartment diffusion, bacterial mRNA, membrane receptors) and provides alpha estimates consistent with prior analyses.
• Training with model-specific datasets can reduce misclassification between closely related models (e.g., CTRW vs ATTM).