Agenda

Round-table update on analysis for the Predictive processing data release manuscript

Please submit 2 slides maximum prior to the meeting at the discussion link at the bottom of this meeting page. Any analysis is welcome, including from new participants.

At the onset of the meeting, we will review the number of presentations to go over as well as introduce any new comer to the meeting.

Meeting Recording

.

Meeting Notes

LFP Oddball Oscillation And Cell-Type Analysis: Antonio presented LFP evidence for a stimulus-specific 25–35 Hz beta/low-gamma response in visual cortex, with increased tagged-neuron firing and spike–field coupling during oddballs; Jerome, Alexander, Samuel, and Sarah proposed analyses to test locality, firing-rate independence, and fast-spiking/PV involvement.

Oscillatory Response: Antonio showed example LFP traces, wavelet time-frequency analyses, Hilbert-transformed 25–35 Hz activity, and PSD measures. Standard stimuli produced increased power, orientation oddballs produced a stronger increase, omissions produced no increase or a decrease, and standard versus hold stimuli were not significantly different in the summarized analysis.

Analysis Scope: Antonio clarified that panels A and B were illustrative data from a single electrode, while panels C–E pooled electrodes and animals. The showcase focused on visual areas V1/V2–V3, particularly layers 2 and 3, where the effect was strongest; related effects appeared across layers, while prefrontal recordings did not show an equally strong LFP effect. Antonio agreed to relabel the timing axis to make clear that the response was centered rather than aligned to stimulus onset.

Tagged Neurons: Antonio selected optotagged neurons using both response magnitude and response latency, initially showing an example from one animal/session and then reporting a larger pooled set of more than 100 tagged neurons from visual cortex layers 2 and 3. Non-tagged units did not show a clear oddball firing-rate increase, whereas tagged units did.

Spike Field Coupling: Antonio reported that example tagged and non-tagged neurons tended to spike near the trough of the oscillation and that spike–field coherence increased during oddballs for both groups, even though firing rate increased mainly in tagged units. Antonio described this as a correlation consistent with possible synchronization of other units by somatostatin-positive neurons, while noting that additional analysis is required.

Follow-Up Analyses: Jerome suggested examining fast-spiking units because low gamma is commonly associated with PV-interneuron interactions. Sarah offered to combine her waveform-classification code with Antonio’s optotagging analysis, and Antonio agreed to investigate waveform-based PV candidates and publish the analysis code on GitHub. Samuel also suggested testing, within individual neurons, whether PPC remains consistent across oddball types independently of firing-rate changes. Alexander recommended current-source-density analysis by taking the second spatial derivative of voltage across contacts to reduce far-field contamination, while cautioning that differentiation amplifies noise; Antonio agreed to investigate this and prepare an interactive LFP viewer.

Integrated Data Analysis Platform Tim presented a local web-hosted Python application that combines project datasets, QC workflows, analyses, and reproducible reports, and Jerome asked the team to test it through a shared repository and resource discussion.

Platform Purpose: Tim explained that the project spans multiple repositories, contributors, modalities, cohorts, and analysis methods, making it difficult to track the full workflow. His application provides a unified interface for selecting datasets, subjects, sessions, QC procedures, and analyses instead of repeatedly modifying scripts.

Workflow Design: Users can create custom cohorts, select sessions, choose QC and analysis modules, and obtain combined outputs in the application or exportable PDF reports. The application uses plug-in modules inside a harness that identifies required files, downloads missing assets, runs the selected analysis, and records outputs and execution provenance.

Current Coverage: Tim reported that the platform already integrates QC scripts, receptive-field analysis, orientation/direction tuning, and oddball analysis for mesoscope sessions, with Neuropixels/ecephys QC integration underway. The shared overview described coverage of mesoscope QC, validated receptive-field analysis, orientation/direction tuning, signal quality and SNR, ROI and motion-related QC, and reviewed oddball results.

Repository Coordination: Tim requested a pinned discussion listing collaborators’ repositories, notebooks, scripts, and Code Ocean capsules with brief descriptions. Jerome supported creating a discussion thread and asked the team to try the platform and provide feedback; Tim agreed to post the repository link and work with contributors to verify that each integrated module behaves as intended.

Electrophysiology QC And Waveform Classification: Sarah presented waveform-based identification of putative fast-spiking cells and an interactive QC dashboard, while the group discussed anatomical uncertainty, CCF boundary errors, functional alignment, and an iterative release strategy.

Fast-Spiking Classification: Sarah and Marina applied WaveMAP to QC-passing units from thalamus, hippocampus, motor cortex, visual cortex, striatum, and prefrontal cortex. Fast waveform units formed a distinct cluster in some areas and were treated as putative interneurons; Sarah noted that a firing-rate threshold, such as above 10 Hz baseline, could further narrow candidates for putative PV cells.

Classification Criteria: Sarah reported that fast-spiking units are not uniformly high firing: applying the 10 Hz threshold leaves a more selective group in motor and visual cortex, raising the question of whether lower-firing fast waveforms represent another cell type or whether the threshold excludes some PV cells. She proposed integrating waveform clusters into the QC tool so users can select cell types for downstream oddball or other analyses.

Interactive QC Dashboard: Sarah demonstrated an HTML dashboard for selecting cohorts, mice, probes, broad brain structures, and QC-passing units, then examining unit yield, depth profiles, session time series, and custom relationships between metrics. Current QC filters included strict ISI violations, presence ratio, amplitude cutoff, SNR, and exact Allen CCF anatomy; Sarah said adjustable thresholds could be added later.

Anatomical Mapping Limits: Stefan noted that neurons appearing in fiber-track labels are likely due to anatomical mismatches and that CCF assignments can have approximately 80–100 microns of error, potentially shifting a unit across cortical layers or into white matter. Sarah and Stefan agreed that major brain-region assignments should still be provided with explicit confidence or boundary caveats, rather than withholding all labels.

Functional Alignment: Stefan described current-source-density signatures as a possible way to refine layer boundaries, especially in visual cortex, while Sarah noted that comparable reference signatures for prefrontal and subcortical regions are uncertain. The group discussed reporting accuracy as a function of distance from known borders and potentially excluding or flagging units near uncertain boundaries.

Iterative Release: Jerome recommended releasing an early version of the dashboard and iterating rather than waiting for a fully optimized anatomical solution. Jerome also suggested realigning individual probes and exploring additional metrics, such as EI ratios or frequency-based measures, to validate or refine anatomical assignments.

Custom Exploration: Sarah showed a custom explorer allowing users to compare arbitrary metrics, such as SNR versus depth or presence ratio versus repolarization slope. She presented it as an exploratory way to understand the breadth and redundancy of available unit metrics, rather than as a tool limited to predefined scientific hypotheses.

Running-State Effects On Oddball Responses: Roberto confirmed that weak Neuropixels motor-oddball responses are associated with few oddballs occurring during running, leading Jerome and Roberto to require running-conditioned analyses and add a compact running-state panel to the results.

Running-State Check: Following Jerome’s suggestion, Roberto calculated the percentage of oddballs occurring while animals were running, defined as wheel speed above 1 cm/s at oddball onset. In the Neuropixels motor cohort, most animals were running during fewer than 20% of oddballs, with only two animals showing frequent running.

Modality Comparison: Roberto found more running during oddballs in the other datasets, including mesoscope, than in Neuropixels. Jerome noted that the mesoscope data therefore likely contain more observable halt responses, while the Neuropixels dataset is intrinsically more difficult to analyze for this effect.

Response Relationship: Roberto compared oddball response amplitude with the percentage of oddballs delivered during running. The result supported the explanation that the two animals with clearer motor halt responses were also the animals that ran during oddballs; the displayed cohort-specific regressions included both nonsignificant motor relationships and stronger negative sequence relationships.

Analysis Decision: Jerome stated that halt analyses must be conditioned on running behavior because combining running and non-running trials could be misleading. Roberto agreed and will prepare a more compact figure, likely as small panels complementing the existing running-analysis figure.

Interactive Figure Hosting: Sarah raised a storage and deployment problem because her 40 MB interactive HTML dashboard could not be uploaded to GitHub, and Jerome agreed to investigate shared-asset hosting options for the project.

File-Size Constraint: Sarah explained that the dashboard is generated by downloading the required data and embedding it into an approximately 40 MB HTML file, which was rejected by GitHub. She asked whether modalities should be split or whether the display should instead use a smaller centralized table of key metrics.

Hosting Follow-Up: Jerome said he would discuss the issue with others and investigate the ongoing effort to store derived assets in shared space. Jerome noted that the project’s increasingly rich visualizations should not be constrained prematurely by repository size limits and committed to finding a practical solution for everyone.

Upcoming Presentation Schedule: The team agreed to continue the work in a future meeting, with Alexander’s receptive-field update likely deferred until the meeting after next because this week’s Friday meeting is being skipped.

Receptive-Field Update: Alexander said the receptive-field work is progressing, including replication of AI code with human code, but requested another two or three weeks before presenting a figure. When Jerome asked about presenting the following week, Alexander said it was probably too early because the Friday meeting was being skipped.

Next Meeting: Jerome said he would send the next meeting invitation and that the team would meet if there were material to share; otherwise, the meeting might also be skipped.

💬 Join the discussion for this page on GitHub