Discuss Ecphys QC Figure panels

[Sarruedi]

Yuheng (@yhcheah804) and I started to create a series of plots for Figure 7, panel E focusing on the QC metrics for single unit data. The panel's overall logic is to give readers an at-a-glance sense of unit quality and, importantly, its stability across sessions and probes before any further analyses or conclusions are drawn.

In the first plot we focused on the SNR metrics provide for all units to illustrate the distribution of SNR vallues across all units in a given session. The first plot summarizes the signal-to-noise ratio (SNR) reported for every unit, shown as a cumulative distribution so the whole shape of the SNR distribution in a session is visible (not just a mean).

We plot this separately for each cohort and overlay the first recording session against the last:
Motor cohort: first (Sensorimotor) vs. last (Duration) session
Sequence cohort: first (Sequential) vs. last (Sensorimotor) session

The point is to check whether unit quality holds up over the course of the experiment. If the last-session curve sat well to the left of the first, that would potentially indicate some degradation in the quality of the actue recordings tissue response, drift, etc.). Overall, in the current data the first/last curves sit close together (thought the first sessions have slightly better SNR), which is the reassuring result we wanted to document.

Plot 1 - Overview - All Units:

Plot 2 All animals, first versus last session across probes A, B, C:

The same first-vs-last SNR comparison, but broken out by probe, with each session drawn as its own thin trace. This lets us see whether the pooled stability in plot 1 is uniform or whether a particular probe/track is driving it, and whether any single probe shows session-to-session spread larger than the rest. It also helps to justify what cut off to use to identify 'good units' for further analysis.

Plot 2 All animals, first versus last session across probes D, E, F

Plot 3: SNR as a function of cortical depth (probes A–F)
For each probe we plot SNR vs. cortical depth (µm) as a scatter across all units (x axis SNR). This shows where along the electrode track well-isolated units are enriched, and how their quality varies with depth; i.e. which portions of the probe are contributing the high-SNR units versus noise-dominated regions. This ties unit quality back to anatomy along the track.

We'll next extend panel E with the standard complementary single-unit QC metrics, each with the same first-vs-last / per-probe / vs-depth logic where it's informative:

Refractory-period (ISI) violations metrics: contamination / isolation quality
Firing rate: including firing rate over time: to catch units that drop out or appear over the course of the recording
Amplitude over time: to catch drift and unit instability
(and likely presence ratio / amplitude cutoff if space allows)

Happy to take input on which additional metrics are highest priority and on thresholds we should annotate on the plots. @jeromelecoq

@severine2305 It would be fantastic to know which areas are targeted by probe A, B, ... and if possible we can try to integrate layer- and depth-dependent anatomical labels at some point. We were wondering how to handle units with negative depth, as we haven't implemented any selection criteria yet. We wonder if there may still be artefactual units/noise included.

[severine2305]

[Sarruedi]

Great, thanks so much!

[severine2305]

area alignment in CCF is done now, you should get it very soon. Negative depth units must be noise

[Sarruedi]

Ok, so we can exclude units with negative depths moving forward and add additional layers of selection criteria to refine the pool of considered units.

[severine2305]

Hi Sarah, in fact I don't know why you get negative depth units, I need to investigate

[Sarruedi]

Hi Severine, it would also be good to clarify the range, is 0 to top of the brain or the tip of the probe in the current reference frame?

[severine2305]

In all my plots 0 is the tip, it will become obvious once you look at the ccf aligment. @rcpeene is the ccf shared in the NWB now?

[Sarruedi]

Cumulative Distributions for good units only for each mouse (units pooled across all 6 probes)

Cumulative Distributions for good units only for each mouse (units for each of the 6 probes)

[maierav]

Super! Only suggestion that comes to mind is to maybe add the absolute unit counts to the legend (n = ...). This way, a reader gets both the relative (ratio/fraction) counts as well as the absolute counts, which might be especially interesting for the Probes.

[Sarruedi]

For now, we reverted the depth axis so 0 is indeed at the bottom of the depth plots.

Here's an illustration figure panel to show the firing rates of good units as a function of depth (binned at 100um) resolution. Each animal is a column in the heatmap to give a quick overview of the depth profile before we do any potential further alignment to CCF of functional signatures as CSD.

"Motor: Sensorimotor Context" - FR by Depth plots 100um depth resolution

[mariannasidiropoulou]

Hi everyone! Together with Sarah (@Sarruedi) and Yuheng (@yhcheah804), we’ve compiled our current Ecephys analyses into four slides (across the three of us) for tomorrow’s meeting.

These analyses focus on the ecephys dataset and are intended to contribute towards Figure 7E and Figure 7I of the data-release manuscript. For the QC side, we’ve been looking at single-unit recording quality across mice, probes and sessions, including the SNR analyses discussed above and firing-rate profiles along the acute probe insertions.

We’ve also started exploring the waveform diversity of the QC-passing units using WaveMAP, looking at the extracellular waveform landscape and waveform classes, their electrophysiological properties, and how these classes are sampled across the different anatomical areas in the dataset.

I am attaching the four-slide summary here. We’d be very happy to present them at tomorrow’s meeting and discuss the analyses with everyone. Looking forward to hearing everyone’s thoughts! : )
Data_release_summary_4slides_RuedigerLab_Ephys.1.pptx