Why use quality metrics?

See this talk at the annual Neuropixels course about quality control.

Advances in silicon manufacturing technology have led to high-throughput devices (e.g. Neuropixels probes) that now allow us to record hundreds of neurons simultaneously. These new methods generate unprecedented amounts of data, and thus new software methods are needed to process this data. In particular, because we are recording the spikes of hundreds of neurons at the same time, we need to group together spikes that are similar; they are likely to come from the same neuron.

Novel automated methods have been developed to do so, but the output of these methods is not perfect and it requires manual assessment to fine-tune the results and get rid of artifacts. Typically, this manual curation:

• is laborious and slow (typically requires several hours per dataset)
• is hard to standardize and reproduce: different manual curators typically have vastly different assessments, and the even same manual curator can make different curation choices on the same dataset!
• needs to be done all over again if any settings change or a new spike sorting algorithm comes out.

Using bombcell's automated quality metrics resolves all these problems:

• It is fast (less than 10' per dataset for a first run, and less than 3' for subsequent runs on a typical 1 hour dataset and a typical computer WITH 32GB of RAM)
• It is standardized and reproducible, and it is additionally very easy to modify classification thresholds. Additionally, using automatic quality metrics allows us to easily vary curation parameters and see how they affect our scientific results, for instance by plotting the quantity we are measuring as a function of each quality metric (see Fig. 2 Harris et al., Nat. Neuro, 2016). This allows us to choose thresholds in a scientific rather than in an empirical way.
• If any settings change or the data needs to be spike sorted again differently, bombcell can simply be quickly run again!

Optionally, you can also use bombcell's metrics to aid in your manual classification.