Nonnegative matrix factorization for analyzing state dependent neuronal network dynamics in calcium recordings.

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Calcium imaging enables high-throughput recording of neuronal activity in vivo, capturing single-cell resolution.
• Analyzing the multidimensional nature of calcium imaging data presents significant challenges.
• Traditional statistical methods often average neuronal responses, losing critical temporal dynamics and relative activity patterns.

Purpose of the Study:

• To adapt and evaluate Nonnegative Matrix Factorization (NMF) as a dimensionality reduction (DR) technique for calcium imaging data analysis.
• To compare the performance of NMF against alternative DR methods using both simulated and in vivo neural recordings.

Main Methods:

• Dimensionality Reduction (DR) techniques were applied to calcium imaging datasets.
• Nonnegative Matrix Factorization (NMF), a DR method with positivity and linearity constraints, was specifically adapted.
• NMF performance was benchmarked against other DR methods using artificial and in vivo neuronal activity data.

Main Results:

• NMF accurately captures the underlying dynamics present in calcium imaging recordings.
• The adapted NMF method demonstrated superior performance compared to commonly used alternative DR techniques.
• NMF effectively reduces data dimensions while preserving essential variance and temporal information.

Conclusions:

• Nonnegative Matrix Factorization is a well-suited and powerful tool for analyzing complex calcium imaging data.
• NMF offers a more comprehensive analysis of neuronal activity, preserving temporal dynamics lost in traditional methods.
• This study highlights NMF's potential to advance the interpretation of in vivo neural recordings.