Spatiotemporal feature learning for actin dynamics.

Area of Science:

• Cellular and Molecular Biology
• Biophysics
• Computational Biology

Background:

• Dictyostelium discoideum is a model organism for studying cell motility and pattern formation.
• Cell migration relies on actin cytoskeleton dynamics, sensitive to microenvironmental cues like stiffness, topography, and electric fields.
• Understanding how microenvironmental factors influence cell behavior is crucial for biological research.

Purpose of the Study:

• To investigate if machine learning can infer microenvironmental conditions (electric fields, nano-topography) from actin wave videos in Dictyostelium discoideum.
• To identify visual features of actin waves that correlate with specific microenvironmental characteristics.
• To develop computational methods for analyzing cell dynamics in response to physical cues.

Main Methods:

• Utilized three machine learning techniques: dictionary learning, scattering transforms, and optical flow.
• Analyzed video microscopy data of Dictyostelium discoideum actin waves.
• Developed frame-by-frame prediction models to classify microenvironment types based on actin wave patterns.

Main Results:

• Dictionary learning and scattering transforms effectively classified cells based on nano-topography by analyzing static image features.
• Optical flow, by tracking stable cellular features over time, proved effective in predicting the presence of external electric fields.
• The study demonstrated that distinct actin wave patterns correlate with different microenvironmental conditions.

Conclusions:

• Machine learning analysis of actin waves provides a robust method for inferring microenvironmental properties.
• Specific machine learning approaches are better suited for identifying different types of physical cues.
• This computational framework can be applied to study collective cell dynamics in various biological systems using video microscopy.