Written by: John Sherwood, Director of Neuroscience, and Brooke Yee, Operations Manager
Background
One way that neurons communicate is via all or nothing physiological events called action potentials. Interestingly, action potentials can be readily and reliably recorded from outside of the neuron by electrodes placed within 100 μm of a neurons cell body. These extracellularly recorded action potentials are fast, typically lasting on the order of just a few ms. It is this fast time-course, and their sharp shape, that has led extracellularly recorded action potentials to be commonly referred to as ‘spikes’.
The Goal
An important question of the day is how populations of neurons and their associated neuronal networks encode and process information in the brain. One might argue that the best way to approach this question is to record from as many neurons as humanly possible. In light of what we know about spikes, a classical electrophysiology approach to achieving this goal has been to implant as many extracellular electrodes as possible. LeafLabs and the Boyden Lab have been at the vanguard of this effort, developing high-density silicon probes and data acquisition system capable of recording from 1000’s of implanted electrodes simultaneously project Willow.
The Problem
An outstanding question is “How do we assign each spike to the correct neuron?”. A task referred to as ‘spike sorting’. Presumably, errors made early in the analysis chain will propagate downstream and impact our ability to interpret data. Much like taking a room full of coherent conversations, translating words between those conversations, and then expecting to be able to extract meaning from the incoherent mess. Historically, the process of spike sorting has largely been a cottage industry; this is reflected by the growing number of algorithms being developed to tackle this problem. With all algorithms requiring some level of human curation. A major problem is that very little ground truth data exists to test how accurately any given algorithm, and its user, are performing.
The Science
In a recently published and heroic study, the Boyden Lab developed a system to perform robot assisted in vivo patch clamp recordings (single neuron) paired with simultaneous co-localized multi-unit recordings, collected using extracellular high density silicon probes. As the patch clamp recordings report the activity of only a single neuron, they can be used as ground-truth data for validating spike clusters sorted from the multi-unit extracellular recordings. This much needed data set is invaluable for characterizing the performance of spike sorting algorithms. Deeper analysis of these data, provides evidence to support a long held belief that high density probes, and spatially oversampling, improves the performance of spike sorting algorithms; as does increasing the number of electrodes at a fixed density. Put simply, multimodal electrophysiological recordings will be instrumental in the validation of spike sorting algorithms, and the future optimization of silicon probe design (electrode number and geometry).
Three of the papers authors are members of the LeafLabs team; Chris Chronopoulos, Charlie Lamantia, and our former Director of Neuroscience Justin P. Kinney.
Read the full article in the Journal of Neurophysiology here!
