Agenda

Presentation on mesoscope QC data analysis for the data release manuscript. First discussion with preliminary plots from @davisgrubin

Meeting Recording

Meeting Notes

Mesoscope Data Quality Metrics Discussion: Davis, Jerome, Farzaneh Najafi, Marcel, and Lucas discussed proposed metrics for evaluating mesoscope data quality, including the sustained exceptional event score, robust SNR, and the Allen event metric, with input from Dr. Najafi on their origins and applicability; the team explored how these metrics could be used to assess data quality across sessions and modalities, and considered their strengths, limitations, and potential for identifying good and bad neurons.

Exceptional Event Score Explanation: Davis explained the sustained exceptional event score, which identifies at least one highly unlikely event in a trace by calculating the half sample mode as a baseline, then summing negative log probabilities of positive deviations above the mode; Farzaneh Najafi added context about its development and use in their postdoc and referenced its inclusion in the Cayman paper.

Robust SNR Metric Description: Davis described the robust SNR metric as the difference between the 95th and 50th percentiles of the DFF trace, with Gaussian smoothing applied to obtain residuals and the median absolute deviation used for noise estimation, making the metric more robust to outliers; this metric provides an aggregate statistic over the whole recording.

Allen Event Metric and Gaussian Fit: Davis detailed the Allen event metric, which tests how closely detected events fit an exponential distribution and whether noise outside events is Gaussian, using the Kolmogorov-Smirnov test to compare observed and theoretical distributions; Farzaneh Najafi and Marcel clarified the interpretation of the score and its assumptions.

Metric Application and Visualization: The team reviewed plots showing metric distributions across ROIs and planes, discussed how to use thresholds to classify neurons, and considered the impact of depth and session variability; Davis demonstrated binning metrics and visualizing DFF traces for low, medium, and high bins, while Farzaneh Najafi suggested further analysis of outliers and correlation between metrics.

Signal and Noise Separation in Metrics: Jerome and Farzaneh Najafi debated the value of separating signal and noise measures within each metric, proposing that breaking down metrics could help classify units into buckets for analysis, and discussed the trade-offs of increasing metric dimensionality versus simplifying thresholds for data release.

Action Plan for Neuron Classification and Threshold Setting: Jerome, Farzaneh Najafi, Davis, and Lucas outlined next steps for classifying neurons using the discussed metrics, including generating average calcium transients, defining buckets based on signal and noise, and setting consistent thresholds across sessions, with plans to revisit the analysis and figure proposal in the following meeting.

Average Calcium Transient Proposal: Farzaneh Najafi suggested using event detection to generate average calcium transients for each neuron as a bulk measure of quality, then binning neurons by metric thresholds and evaluating the average transients in each bin to identify good and bad neurons.

Threshold Consistency and Manual Validation: The group discussed the importance of applying the same thresholds across all sessions for consistent data quality assessment, and considered manual validation of neuron quality in a subset of sessions to establish ground truth for threshold setting.

Bucket Definition and Classification: Jerome proposed classifying units into buckets based on combinations of signal, noise, and SNR values, and counting the number of units in each bucket across sessions; Davis agreed to take a first pass at defining these buckets based on the data.

Next Steps and Figure Proposal: Jerome and Farzaneh Najafi agreed to revisit the analysis next week, with Jerome planning to draft a proposal for figure 4 that would illustrate metric definitions, example cell types, and bucket distributions across sessions.

Feedback and External Metric Suggestions: Lucas, Jerome, and Farzaneh Najafi discussed the definition of signal in the context of the metrics, considered event detection as a method for signal identification, and referenced external sources and toolboxes for potential metric improvements, including suggestions to consult Peter Rupach and review other SNR metrics.

Signal Definition Clarification: Lucas asked for clarification on how signal is defined in the analysis, with Davis explaining the calculation for the robust SNR metric and Farzaneh Najafi supporting event detection as a method for identifying signal.

External Metric References: Jerome and Lucas discussed consulting Peter Rupach for metric advice and referenced the Cascade toolbox, though noted it was not currently in use; Farzaneh Najafi encouraged sharing any other SNR metrics seen in literature for consideration.

💬 Join the discussion for this page on GitHub