Orientation tuning in mouse V1, one click at a time
Real Neuropixels data. This starts with the figure everyone
publishes — the average response across all recorded neurons, one trace
per grating direction — and lets you click down to the data it was
computed from.
The population average is almost flat. That is not a bug and not a
dead recording: three quarters of these neurons are strongly
orientation-tuned, but they prefer different orientations, so
averaging across them cancels the tuning out. You cannot see that from the
top panel. You can see it immediately one level down.
loading ~2.3 MB of spike times and binning them…
On real data these matter more than on simulated data. Many
V1 units have baselines near zero, so divide by baseline on the TRIAL
selector — a denominator estimated from a couple of spikes —
blows up in a way the per-NEURON version does not. Try
peak = 1 on the NEURON selector to see the tuning survive
averaging.
Click a trace — or, easier with eight overlapping traces, click its legend entry.