Efficient and reliable spike sorting from neural recordings with UMAP-based unsupervised nonlinear dimensionality...

Area of Science:

• Neuroscience
• Computational Biology
• Signal Processing

Background:

• Spike sorting is essential for analyzing extracellular electrophysiology data.
• Dimensionality reduction is a critical step in spike sorting pipelines for effective clustering.
• Current methods often rely on linear or supervised nonlinear techniques.

Purpose of the Study:

• To evaluate the effectiveness of Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction in spike sorting.
• To demonstrate UMAP's ability to improve the performance, efficiency, robustness, and scalability of spike sorting pipelines.
• To compare UMAP against traditional dimensionality reduction methods used in electrophysiology.

Main Methods:

• Application of Uniform Manifold Approximation and Projection (UMAP), an unsupervised nonlinear dimensionality reduction technique.
• Testing on both synthetic and experimental extracellular electrophysiological recording data.
• Comparison of UMAP-based dimensionality reduction with existing linear and ad hoc supervised methods.

Main Results:

• UMAP significantly improved the performance, efficiency, robustness, and scalability of spike sorting.
• The number of correctly sorted neurons increased substantially using UMAP.
• Identification of sparsely firing neurons became more reliable, enabling deeper neural code analysis.
• UMAP facilitates more efficient and automatable spike sorting for large-scale neural recordings.

Conclusions:

• UMAP offers a powerful, unsupervised approach to dimensionality reduction for spike sorting.
• Replacing traditional methods with UMAP enhances the accuracy and reliability of neural data analysis.
• UMAP paves the way for more advanced and scalable spike sorting pipelines for high-density neural recordings.