Session Kit: Module 18: Data Cleaning and Preprocessing

Everything needed to run Module 18 as a taught session: prep, timing, materials, misconceptions, rubric.

Generated from modules/module18.md. Edit the module page, not this file.

At a glance

   
Duration 4-5 hours
Capability target Produce a reproducible preprocessing release that transforms raw or intermediate connectomics outputs into analysis-ready data, with explicit quality gates and full provenance. Students will be able to identify the specific cleaning operations that shape biological conclusions, justify every threshold decision, and document their preprocessing pipeline so that another researcher can audit and reproduce it.
Learners leave with Preprocessing decision table (one row per issue, columns: issue, policy, threshold, rationale, impact)

Before you walk in

Learners should arrive having covered:

Materials

Run of show

Time Segment Your note
00:00-08:00 Setup and target framing  
08:00-18:00 Instructor modeling: ingest and anomaly screening  
18:00-32:00 Team preprocessing design  
32:00-44:00 QC pass  
44:00-54:00 Cross-team review  
54:00-60:00 Competency checkpoint  

The activity

Scenario: Your team receives a connectomics export from MICrONS minnie65 (CAVE materialization v795) containing: a synapse table (4.2 million rows) with confidence scores, a segment table (120,000 segments) with volumes, and a cell-type annotation table (8,400 classified neurons). Initial inspection reveals: 12% of synapses have confidence scores below 30, 35,000 segments have fewer than 2 synapses, 847 segments intersect the volume bounding box, and 23 segment IDs appear in the synapse table but not in the segment table.

  1. Artifact triage: classify each issue (low-confidence synapses, small segments, boundary neurons, orphan IDs) by likely biological impact and propose a cleaning policy for each.
  2. Threshold justification: for synapse confidence and segment size thresholds, propose two candidate values each and argue for your preferred choice. Explain what biological signal you might lose at each threshold.
  3. Implement preprocessing pipeline: write pseudocode or notebook-level steps for the full cleaning workflow, from ingest through release.
  4. QC comparison: compute (or estimate) pre/post metrics: total synapse count, total segment count, mean degree, graph density, and the fraction of each cell type remaining after cleaning.
  5. Release note: produce a one-page release note that includes: input dataset version, all thresholds and parameters, code reference, QC metrics with pass/fail calls, and known residual risks (e.g., “boundary neurons were excluded, which may underrepresent connectivity of neurons near volume edges”).

What learners hand in

Misconceptions to target

These are the errors this session exists to prevent. Surface them in the debrief rather than pre-empting them in the lecture — a misconception a learner has voiced is far easier to correct than one they are holding silently.

Naming the norm

Every session is a chance to make one piece of the hidden curriculum explicit. Pick a moment where you would normally just do the professional thing, and say out loud why you are doing it — then ask whether anyone was taught that.

For this session, the candidate is whichever norm the activity most depends on: stating an assumption in the same sentence as the claim, recording the version a number came from, or saying “uncertain” and having it count as a real answer. See the hidden curriculum for the collected set and why naming them is a fairness intervention rather than etiquette.

Assessment

Grade the reasoning, not the answer. A correct call with no evidence chain should not outscore a well-reasoned incorrect one — and saying so publicly changes behaviour within one session.

Exit prompt

Take one connectomics table (real or mock) and write:

  1. Three cleaning rules with rationale tied to specific data artifacts.
  2. Two QC thresholds with associated pass/fail actions and biological justification.
  3. One sensitivity analysis: what happens to your key metric if you relax or tighten your primary threshold by 20%?
  4. One limitation that remains after preprocessing, stated concretely enough to guide interpretation.

If this session goes wrong


All session kits · Facilitator guide