• Date: 2026-07-21
• Time: 09:00AM (PT)
• Location: Teams Meeting

Agenda

Follow-up discussion on mesoscope QC data analysis for the data release manuscript. Preliminary plots from @davisgrubin. Figure scaffold is available here

Meeting Recording

Meeting Notes

Event Detection and Background Definition in Mesoscope Data: Davis presented the methodology for defining events and background in mesoscope calcium imaging data, with input from Jerome, Farzaneh, Nicholas, and Sarah, focusing on the use of Allen/OASIS-style deconvolution and the calculation of background noise to assess signal quality.

Event Window Definition: Davis explained that events are defined using the Allen/OASIS-style deconvolution algorithm, with event windows set as 0.5 seconds before and 2 seconds after each detected event. When events occur within 0.5 seconds of each other, they are merged into a single window, taking the midpoint as the event time. This approach was discussed and clarified with Jerome and Nicholas, who asked about the rationale and implications for temporal precision.

Background Calculation Method: The background is calculated by excluding event windows from the trace and using the remaining portions to estimate noise, specifically via the median absolute deviation. Davis described this process and addressed questions from Farzaneh and Sarah about the distinction between event and background, emphasizing that the background is defined as non-event portions of the trace.

Signal-to-Noise Metric: Davis used the calculated background to assess the amplitude of detected events, categorizing them by how many standard deviations they are above the background. Events are grouped into less than two, two to four, and greater than four standard deviations, which forms the basis for subsequent clustering and quality assessment.

Algorithmic Assumptions and Limitations: Nicholas and Sarah raised questions about the assumptions underlying the event merging algorithm and the potential for spurious event detection in noisy traces. Davis acknowledged these limitations and described the heuristic nature of the approach, noting that the method aims to distinguish genuine calcium transients from random fluctuations.

Clustering and Quality Control Metrics for Calcium Imaging: Davis demonstrated the use of k-means clustering to categorize ROIs based on event amplitude relative to background, with discussion from Jerome, Farzaneh, and Sarah about the interpretation of clusters, the influence of cell type and depth, and the potential for additional metrics.

K-Means Clustering Approach: Davis applied k-means clustering to group ROIs based on the proportion of events in each standard deviation category, resulting in clusters that reflect different signal qualities. The clusters were visualized and discussed, with Jerome and Farzaneh asking about the interpretation and the relationship to cell types and anatomical layers.

Depth and Cell Type Effects: Jerome explained that the depth of imaging planes and the mixture of cell types (inhibitory and excitatory) influence the observed signal characteristics, with superficial layers showing dendritic signals and deeper layers showing somatic signals. Davis and Farzaneh discussed how these factors affect the clustering and signal-to-noise metrics.

ROI Size and Quality Assessment: Sarah suggested plotting signal-to-noise metrics against ROI size to assess whether larger (somatic) ROIs are more reliable than smaller (dendritic) ones. Davis noted that area metrics were not yet implemented but agreed this would be a valuable addition, and Jerome mentioned the availability of classifier outputs to distinguish somatic ROIs.

Metric Interpretation and Figure Design: Farzaneh and Jerome discussed how the chosen metric (proportion of high-amplitude events) relates to activity and quality, noting that a neuron may have good SNR but low activity. Davis clarified the metric's calculation, and Sarah recommended showing the distribution of standard deviations across ROIs for better intuition.

Feedback and Iteration on Figure Design for Data Release Paper: The team, led by Jerome and Sarah, provided feedback on Davis's draft figures, emphasizing the need for interactive visualization, inclusion of multiple metrics, and clear representation of depth and ROI characteristics, with plans for further iteration and integration of suggestions.

Interactive Visualization Proposal: Sarah proposed that the figure panel include interactive elements, allowing users to explore the data and select ROIs of interest. Davis described the current implementation, which was primarily for debugging, and Jerome suggested including this as a supplementary asset in the manuscript.

Figure Panel Structure and Metrics: Jerome outlined the intended structure of the figure panel: the top row shows ROI extraction and event detection, the second row explains the derived metrics, and the bottom row displays class distributions across areas and contexts. The team discussed whether to use a single metric or multiple complementary metrics.

Depth Representation and Metric Iteration: Jerome and Davis discussed the importance of highlighting depth in the figure, noting patterns in signal quality across imaging planes. Davis was encouraged to iterate the figure using feedback, including trying alternative metrics such as lower percentile background measures as suggested by Sarah.

Clarification of Metric Calculation: Farzaneh requested clarification on whether the metric represents the number of events or their magnitudes, and Davis explained that it is based on the proportion of events in each standard deviation category. Jerome further clarified the calculation and its interpretation.

Plans for Further Analysis and Cross-Modality Application: Lucas asked about applying the same event detection and quality control pipeline to SLEP2 data, and Jerome confirmed its applicability, with the team planning to iterate figures and discuss metrics in future meetings, including cross-team collaboration.

Cross-Modality Pipeline Application: Lucas inquired whether the event detection and quality control pipeline developed for mesoscope data could be applied to SLEP2, and Jerome confirmed that the approach is relevant and could be used for transmission event analysis in SLEP2.

Future Collaboration and Scheduling: Jerome discussed plans to reach out to team members for scheduling future presentations, acknowledging summer vacations, and Sarah confirmed their availability for mid-August. The team agreed to continue iterating figures and discussing metrics in upcoming meetings.