Dashboard
One dose, one compound, and an explicit account of which numbers were measured and which this model invented.
One concentration axis
Insect engagement, vertebrate engagement and circuit response are all dimensionless fractions, so they share one y-axis. No second scale is ever drawn.
Selectivity — values, not a verdict
A ratio is not a safety margin. The vertebrate engagement at this dose is printed beside it, and the receptor that becomes limiting first is named.
Circuit consequence
Vehicle → treated on the committed MaleCNS cut. Rates are model output, not recordings.
Is the effect wiring-dependent?
A separate question from "did the circuit move". Permutation nulls destroy one kind of structure at a time; the headline statistic is the empirical p with its resolution, never a bare z.
What can I trust?
Per layer, not blended. There is no single confidence percentage here on purpose.
Compare compounds
Receptor selectivity beside circuit selectivity. The two do not rank compounds the same way.
Dose-response
Where saturation happens, and why changing an EC50 barely moves a receptor that is already full.
Engagement and circuit response on one axis
Same three series as the dashboard, drawn large. The vertical rules are the current dose and the concentration at which the vertebrate receptor reaches 20% engagement.
Saturation check
Computed from this run: how far a two-fold error in the cited parameter would move engagement at the current dose.
Every receptor in this row
Solid = insect, dashed = vertebrate counterpart. Rows with no sourced value are not drawn: absent is not zero.
Genotype
Wild type against sourced resistance alleles. Shifts are per compound: an allele that moves deltamethrin need not move DDT.
Potency under each allele
Engagement at the current dose, with the sourced fold-shift beside it.
Circuit response under each allele
MN9 firing rate on the same cut, same drive.
Before and after
Per-compound specificity
The same allele applied to a second compound, to show the shift is not a property of the receptor alone.
Mixtures
Two compounds, their targets, and whether the combination beats the null expectation.
Observed against expectation
Single agents and the combination, with the model's expectation drawn as a reference line.
Isobologram
Iso-effect pairs against the Loewe additivity line. Points below the line are more than additive.
Receptors touched by the mixture
Where the two compounds meet, and where only one of them acts.
Evidence
Every number carries a classification. Click any chip to open its provenance.
Classification key
This compound, row by row
The library rows behind the current run, with the parameter type, the source relation and the species.
Fact · Inference · Unknown
Generated for this experiment from flylab/analysis/claims.py.
Claim chain
Result → parameter source → engagement → mechanism rule → expression → MaleCNS edges → transmitters → engine → readout.
Library census
What the whole library is made of, by parameter type and evidence tier.
Rank validation
Model orderings against published ones, with the recorded discrepancies.
Deeper analyses
These cost real compute. Each one states its estimated runtime before it runs, and the browser build defaults to the cheap settings.
Scorecard
The same dose, two receptor panels. Occupancy is a Hill calculation on literature-order EC50s, not a measurement.
Insect vs vertebrate occupancy at the current dose
Paired targets only. Hatched bars are class placeholders: the compound has no sourced value at that receptor.
Selectivity pairs
log10 of the vertebrate/insect EC50 ratio; positive means the insect target is the more potent one.
Every receptor in the library row
Table view of the chart above, including unpaired targets.
Circuit gains
Vehicle is 1.0. Blue = suppressed, red = enhanced.
Mechanism table
Which gain each occupancy touches, and the formula. Rules active at this dose are listed first.
Curves
Occupancy and circuit readout across the dose ladder.
Occupancy vs concentration
Solid = insect target, dashed = vertebrate counterpart. The vertical rule is the current dose.
Circuit dose-response and model IC50
Model-derived, not an animal IC50: it is the dose at which this simulated readout sits halfway between its low- and high-dose plateau.
Circuit
The committed MaleCNS neighborhood, laid out by hop distance from the seeds.
Top changed nodes
Largest vehicle → treated rate change. Click a row to centre that cell.
Sweet and bitter GRN → MN9 paths
taste_motor graph only. Click a path to highlight it in the viewer.
Spikes
Shiu-style LIF on the neighborhood graph. Vehicle and treated are separate runs at the same seed.
Raster — vehicle
Named seed cells first, then the most active cells.
Raster — treated
Same cells, same seed, drug gains applied.
PSTH — MN9
10 ms bins, summed over both MN9 body IDs.
PSTH — DNp01
10 ms bins, summed over both DNp01 body IDs.
Rates
Taste
The reduced control circuit beside the map-extracted path, at the same dose and drive.
reduced_taste_v0
Hand-built directional control for the bitter veto.
MaleCNS taste_motor path
Real labellar GRN body IDs drive the extracted graph; MN9 is read, never driven.
Side by side
Whole CNS
Traced-cell census with predicted transmitters. This is a census model, not a 166k-cell simulation.
Cells by transmitter
MaleCNS v1.0 traced cells, predicted consensus transmitter.
Cells by superclass
Top 12 superclasses.
Excitation / inhibition index
Vehicle against treated. Dimensionless index, not a firing rate.
Gustatory cells by subclass
Where the labellar GRN seeds come from.
Exposure
One compartment, first-order in and out. Every constant is a teaching placeholder.
Haemolymph concentration C(t)
Bateman profile for the selected route.
Occupancy(t)
Instantaneous equilibrium at each time point.
Uncertainty
How much the model moves when the teaching EC50s and the drive move. Not a biological error bar.
Ensemble credible intervals
Point estimate with the 2.5–97.5% percentile interval over replicates.
Sensitivity tornado
One parameter at a time, halved and doubled, around the base run.
Experiment
Build a design, run it, export it. Randomise and blind it to use the bench as a teaching lab.
Design
Heatmap — compound × concentration
Group mean of the first selected readout.
Results
Group mean ± sd over replicates.
Controls
Is the effect bigger than a shuffled network would give, and is it selective?
Null-model panel
Real effect against the null distribution, per shuffling mode.
Selectivity landscape
Receptor selectivity index against circuit selectivity index, for the current compound and a small reference set.
Notebook
The JSON record of the most recent run, warnings first.
Warnings
Provenance
Import a live_lab CSV
FlyLab never generates live-animal numbers. This slot only holds a table you measured.
Session history
The last 50 runs from this browser. Click one to load it back.
Notebook JSON
No run yet.