Density-based Neural Decoding using Spike Localization for Neuropixels Recordings

Neural decoding is essential for understanding the association between neural activity and behavior. A prerequisite for most decoding methods is spike sorting, the assignment of action potentials (or spikes) to individual neurons. Current spike sorting algorithms, however, can be inaccurate and do not properly model uncertainty of spike assignments, therefore discarding information that could potentially improve decoding performance. Recent advances in high-channel-count probes like Neuropixels (NP) and extracellular analysis pipelines allow for extracting a rich set of spike features to directly decode behavioral correlates using unsorted spiking data. To this end, we propose a density-based decoding algorithm that incorporates our uncertainty about spike assignments in the form of parametric distributions of spike features. Our density-based decoding approach allows for retaining maximum information about the recording and for explicit uncertainty quantification of spike assignments. Our approach can also reduce the computational cost of neural decoding by avoiding spike sorting. With applications to electrophysiological and behavioral data from the International Brain Laboratory (IBL), we demonstrate that our density-based decoding algorithm can outperform decoding algorithms based on thresholding (i.e. multi-unit activity) and algorithms which rely on well-isolated, single-unit activity.