Bridging large-scale neuronal recordings and large-scale network models using dimensionality reduction.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• A core challenge in neuroscience is integrating empirical neuronal recordings with theoretical neural network models to elucidate brain mechanisms.
• Historically, this integration relied on analyzing single-neuron activity and pairwise spike train statistics.
• This traditional approach has limitations in capturing the complexity of neural population dynamics.

Purpose of the Study:

• To review recent advancements in connecting neuronal recordings and neural network models.
• To highlight the utility of dimensionality reduction techniques in analyzing multi-dimensional neuronal population activity.
• To demonstrate how this approach facilitates a deeper understanding of brain function.

Main Methods:

• Review of recent studies employing dimensionality reduction on neuronal population data.
• Application of dimensionality reduction to identify the underlying structure in high-dimensional neural activity.
• Comparison of model predictions with experimental data based on population activity structure.

Main Results:

• Dimensionality reduction effectively captures the multi-dimensional structure of neuronal population activity.
• This approach provides a framework for comparing and integrating experimental recordings with computational models.
• The method has been successfully applied to diverse cognitive functions, including working memory, decision making, and motor control.

Conclusions:

• Dimensionality reduction offers a powerful, unified approach to bridge neuronal recordings and network models.
• This technique enables more incisive comparisons and a tighter interplay between theory and experiment in neuroscience.
• Future research can leverage dimensionality reduction for more comprehensive models of brain function.