PART 05 / Intermediate

Drosophila: anatomy, circuits & datasets

How do we locate a fly circuit without confusing anatomy, function, and simulation?

Guided illustrated lesson

What you will learn

Write a provenance-aware neuron dossier and trace a pathway with evidence for each connection.

Before you begin: Connectomics: wiring as data; From neurons to brains

START WITH A QUESTION

Picture this.

Suppose someone hands you a beautiful map labeled 'the fly brain.' Your first question should be: which fly, which part of its nervous system, and which release? Those details are the coordinates that make the map useful.

Cell type
A classification supported by features such as morphology, connectivity or molecular properties.
Perturbation
A deliberate change used to test how a system responds.
Normalization
Rescaling numbers relative to a specified total or reference.

Read for the idea first. Work through the numbers next. Try answering the questions before opening the explanations.

Get your bearings

Drosophila is a fruit-fly genus; the research animal discussed here is Drosophila melanogaster. Its brain and ventral nerve cord form different parts of the nervous system. Descending pathways carry information broadly from brain toward body-related circuitry, while ascending pathways carry information the other way. A brain reconstruction is not automatically a reconstruction of the entire nervous system. Start a dataset notebook with specimen, sex, reconstructed region, release and annotation source. A neuron identifier is an address inside that context.

Follow a visual question, not every cell

The retina, lamina, medulla, lobula and lobula plate are anatomical regions along branching visual pathways. They are not interchangeable layers in a generic neural-network diagram. T4 and T5 populations provide experimentally studied direction-selective signals associated with ON and OFF motion pathways. LPLC2 provides a focused example for studying looming, or image expansion. Begin with one stimulus question—such as expansion versus translation—and trace the evidence relevant to it. You do not need to memorize an atlas before asking a good question.

Make a three-column evidence card

For a chosen cell type, separate anatomy, measured activity and intervention. Anatomy records partners and morphology. Activity records what happened during particular stimuli. An intervention changes a component and measures a consequence. These columns support different conclusions. A cell responding during behavior may be correlated with the behavior; perturbing it provides a stronger route toward testing a causal role, with its own controls. Put proposed robotic uses in a fourth, explicitly hypothetical column.

Selections change the denominator

Imagine a source with 2, 3 and 5 counted contacts to three targets. Their fractions are 0.2, 0.3 and 0.5. If you remove the last target, the remaining fractions become 0.4 and 0.6 after renormalization. The cell did not change; your question changed from 'fraction of all selected contacts' to 'fraction among the retained contacts.' This is the same trap as changing a survey's population while keeping its percentage label. Fractions are useful summaries, but they are not firing probabilities.

SEE THE RELATIONSHIPS

From structure to behavior.

Fly vision: a labelled conceptual diagram
Figure 5.1 · Fly vision. Original educational diagram; synthetic, not a measured result.
The reasoning pathVersioned specimen then Cell-type dossier then Selected subcircuit then Sensitivity analysis1. Versioned specimen2. Cell-type dossier3. Selected subcircuit4. Sensitivity analysis
A learning flow for this part. Follow the assumptions through to an observable result.

MAKE THE IDEA PRECISE

The mathematics.

f_ij = c_ij / sum_k c_ik
Symbols and units
SymbolMeaning
c_ijIllustrative synapse count from i to j
f_ijFraction of counted outgoing synapses assigned to j
kIndex over included targets

Build the equation, step by step

  1. Pick one source cell i and a defined set of targets. Sum its counted contacts to obtain the denominator.
  2. Divide each target count by that sum. When the sum is positive, fractions over the selected targets add to one.
  3. Changing the selected target set changes the fractions. An empty or zero-contact row requires an explicit missing-data policy rather than division by zero.

WORK THROUGH THE NUMBERS

Worked example

A synthetic source has outgoing counts 2, 3 and 5. Their normalized fractions are 0.2, 0.3 and 0.5. They sum to one because of the normalization, not because a biological signaling law was discovered.

A small Python experiment.

Python 3 · Standard library only · Synthetic educational example. Run this in your own notebook or a local file.

counts = [2, 3, 5]
fractions = [value/sum(counts) for value in counts]
print(fractions)
print(sum(fractions))

Read the code as a story

sum(counts) creates the denominator shared by all three divisions. The fractions sum to one by construction. An empty list or a list of all zero counts needs a deliberate policy before division. Reporting 'no counted contacts' is different from inventing a uniform distribution.

Expected output

[0.2, 0.3, 0.5]
1.0

Change one quantity, predict the result, and then run the code again. Explain any difference before adding complexity.

Test your understanding.

Try it: Remove the target with count 5. Explain why renormalizing the remaining targets changes interpretation.

Self-check: Does a normalized contact fraction equal a firing probability?

No. It is a fraction of counted contacts within the selected data, not a measured transfer function. A model can use it as a chosen weight, but that assumption needs testing and comparison against alternatives.

Where this model stops: This edition does not provide an exhaustive cell-type atlas or a newly reconstructed connectome.

A research question: Which functional claims need activity measurements or perturbations beyond connectivity?

Record evidence in your mastery worksheet →

PREDICT / TRY / EXPLAIN

Audit a selected subcircuit

Use paper, a calculator or the Python example. Write your prediction before checking the result.

  1. Use the three synthetic counts and calculate the original fractions.
  2. Drop the target with five contacts. Record both the retained contact fraction of the original total and the renormalized target fractions.
  3. Write a sentence explaining what each denominator means. Add one missing-data question you would ask of a real dataset.
Compare with a worked solution

Five of ten contacts remain, so retained coverage is 50%. Within that retained subset, the targets account for 40% and 60%. Reporting only 40% and 60% hides the missing half. A useful question is whether omitted partners lie outside the reconstructed volume or were excluded by your own filter.

CLOSE YOUR NOTES

Can you explain it back?

These questions test the reasoning, not just a remembered phrase. Say your answer out loud or write a sentence first.

01Can the same numeric ID safely identify a neuron in two different datasets?

Reveal the reasoning for question 1

Only if an explicit, validated mapping establishes that relationship. Numeric equality alone is not biological identity.

02A neuron responds to expansion. What extra comparison would help test selectivity?

Reveal the reasoning for question 2

Compare translation, contraction and suitable brightness controls under a stated stimulus protocol. A response to one stimulus alone does not establish that it prefers that stimulus.

TAKE THE IDEA WITH YOU

Three things to remember.

  • Keep specimen and release beside every biological identifier.
  • Anatomy, activity and intervention answer different questions.
  • Report what a subcircuit selection excludes, not just what remains.

We have a biological context and a wiring language. Next we turn the membrane picture from Part 1 into a numerical simulation you can check by hand.

Continue to Part 6 →

THE NEXT LAYER

Chapter depth map.

This introduction provides a first model. The following chapters are outlines for deeper future editions; their full treatments have not yet been published.

A nervous system in context

5.1 Brain, body and public maps · Outline
  1. 5.1.1 Brain and ventral nerve cord; ascending and descending neurons
  2. 5.1.2 FlyWire and MaleCNS: specimen, scope, version and access
  3. 5.1.3 Brain regions, neuropils, cell types and stable identifiers
  4. 5.1.4 Olfactory pathways, mushroom body and navigation circuits
  5. 5.1.5 Compare datasets without equating IDs across specimens
5.2 The visual system · Outline
  1. 5.2.1 Photoreceptors and retinal sampling
  2. 5.2.2 Lamina, medulla, lobula and lobula plate
  3. 5.2.3 ON/OFF pathways and T4/T5 direction selectivity
  4. 5.2.4 Visual projection neurons and distinct LC families
  5. 5.2.5 Anatomical diagrams with pathway-specific evidence

From cell type to behavior

5.3 Looming and descending pathways · Outline
  1. 5.3.1 LPLC2, LC4 and the evidence for looming-related processing
  2. 5.3.2 DNp01/giant fiber and escape-related pathways
  3. 5.3.3 Descending versus ascending information
  4. 5.3.4 Inputs, outputs, location and uncertainty in cell-type dossiers
  5. 5.3.5 Robotics analogies must be separated from fly function
5.4 Dataset comparison as an experiment · Outline
  1. 5.4.1 Read official release notes and licenses
  2. 5.4.2 Record segmentation and annotation versions
  3. 5.4.3 Reconstruct a small pathway from source queries
  4. 5.4.4 Cross-check morphology and connectivity
  5. 5.4.5 Report unverified functions rather than filling gaps

Read with the original sources.

These readings support the concepts and deeper investigation. The worked numbers and illustrations above are teaching examples.

Official dataset resource

MaleCNS — official release and downloads ↗

The official download page exposes versioned data, including v1.0 resources. Keep specimen, release, filtering and annotation provenance with every derived graph.

Complete source register →

Where next?