Evaluating linear model frameworks for directed predictive influence estimation from hypothalamic calcium imaging.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Calcium imaging generates high-dimensional, noisy neural signals.
• Statistical tools are needed to infer directed relationships and functional connectivity from these signals.
• Existing methods often struggle to capture directed dependencies accurately.

Purpose of the Study:

• To evaluate linear modeling frameworks for estimating functional connectivity from calcium imaging data.
• To benchmark linear regression models against traditional correlation-based approaches.
• To assess the performance of models in estimating directed predictive influence (DPI).

Main Methods:

• Linear regression models (pairwise and multi-predictor) with and without L1 regularization were applied.
• Models were benchmarked using zero-lag regression coefficients to estimate DPI.
• Cross-validation, hold-out testing, and simulations with predefined networks were employed.
• Analyses used z-scored fluorescence signals and MLSpike deconvolution for spike-rate estimates.

Main Results:

• Linear regression models demonstrated effective estimation of DPI from calcium imaging data.
• The models showed consistent inference across different preprocessing methods (z-scoring and spike-rate estimates).
• Cross-dataset validation confirmed the robust generalization of findings across multiple recordings and behavioral states.

Conclusions:

• A transparent, scalable, and interpretable modeling pipeline for estimating DPI from calcium imaging data was established.
• This pipeline provides a benchmark for evaluating statistical approaches to neuronal network inference.
• The study highlights the utility of linear modeling for understanding functional coupling in neuronal networks.