Active learning of neural population dynamics using two-photon holographic optogenetics

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Advanced techniques like two-photon holographic optogenetics and calcium imaging allow precise neural population monitoring and perturbation.
• Studying neural population dynamics is crucial but hindered by the vast number of potential stimulation patterns and time-intensive experiments.

Purpose of the Study:

• To develop efficient algorithmic methods for selecting optimal photostimulation patterns.
• To identify stimulation patterns that best inform dynamical models of neural population activity.

Main Methods:

• Developed an active learning procedure for low-rank regression.
• Utilized a low-rank linear dynamical systems model.
• Applied methods to mouse motor cortex neural population responses to photostimulation.

Main Results:

• Demonstrated the efficacy of the low-rank dynamical systems model.
• Showcased an active learning procedure that leverages low-rank structure to find informative photostimulation patterns.
• Achieved up to a two-fold reduction in data required to reach a target predictive power on both real and synthetic data.

Conclusions:

• The developed active stimulation design method efficiently identifies informative photostimulation patterns for neural population modeling.
• This approach significantly reduces the experimental data needed for accurate dynamical model inference.
• The underlying active learning procedure for low-rank regression may have broader applications.