Fig 1.
Model comparison for whole-brain data.
(A) Number of input cells vs starter cells. Colours indicate different fitted models (blue—linear; yellow—quadratic; grey—power law; green—exponential; purple—growth). (B) Distribution of standardized residuals for the power-law fit. (C) Same as in A, but on log-transformed data. (D) Same as in B, but for the linear fit on log-transformed data.
Table 1.
External datasets are from [13–15, 18, 19, 21, 22, 25, 32–36]. For easier comparison, we show the ΔAICc values, calculated as the difference between the AICc value for each model and the lowest AICc value per dataset. Lower AIC values indicate a better-fit model, and a model with a ΔAIC (the difference between the two AIC values being compared) of more than -2 is considered significantly better than the model it is being compared to.
Table 2.
ΔAICc for log-transformed data.
Table 3.
ΔAICc for residuals distribution.
Fig 2.
Distribution of slope and y-intercept values for different input areas.
(A) Slope vs y-intercept relationship for individual input areas. Colors indicate functional grouping of brain areas (VIS (dark blue): VISp, VISpm, VISam, VISl, VISal; RSP (blue): RSPv, RSPd, RSPagl; Thal. (cyan): LP, LD, AM, LGd; Distal cortex (green): ORB, ACA, AUD, PTLp, TEa, MOs, CLA). Whole-brain data is shown in gray. Black line is a linear fit through all data points except the whole-brain data. Error bars are 95% confidence intervals calculated using residuals resampling. (B) Data from simulations using the probabilistic model over a range of parameters (Ni, represented by the different colours, and p, indicated by different markers).
Table 4.
ΔAICc for untransformed data for individual brain areas.
Table 5.
ΔAICc for log-transformed data for individual brain areas.
Table 6.
ΔAICc for residuals distribution for individual brain areas.
Fig 3.
Effect of degree distribution of input and starter sets on slope and y-intercept values.
Average degree distribution of either the starter (orange) or the input (green) pools were specified using a configuration model and varied separately. log(ni) vs log(ns) relationships of the resulting networks were fit with a linear model to extract y-intercept and slope values.
Fig 4.
Effect of starter number on area input fraction.
(A) Multivariate linear regression between the area input fraction and starter cell number, location or genotype of starter cells. (B) Area input fraction vs starter relationship for four example areas. Dashed lines are linear fits through the data, for the full dataset (grey line), starters < 200 (blue line) or starters > 200 (orange line) (C) Slope of the area input fraction vs ns relationship for low (blue) or high starter numbers (orange).
Table 7.
Intercept values from linear fits of log(ni) vs log(ns) relationships.
Fig 5.
Effect on relative connection probability on area input fraction vs starter relationship.
Using the probabilistic model, we simulated 5 input areas, all with Ni = 100. 100 independent simulations were repeated to assess the effect of p. For each simulation, the connection probability pi for each input area was randomly drawn between 10−4 and 8*10−3. (A) Heatmap showing the combination of connection probabilities used for each simulation. The simulation shown in plots B-D is indicated by a red arrow. (B) ni vs. ns relationship for all input areas for one example simulation. (C) Area input fraction for low starter numbers (lowest 10%) or for high starter numbers (highest 10%). (D) Area input fraction vs ns relationship for each area. Black line is a polynomial fit. (E-F) Area input fraction vs rank of connection probability for low starters (E) or high starters (F). Data from the simulation plotted in B-D are shown in grey. (G) Relationship between the ratio of the area input fraction for high vs low starters, and the normalized connection probability per area. Data from the simulation plotted in B-E are shown in grey.
Fig 6.
Estimation of Ni and p per area in experimental dataset.
(A) Four example areas with input vs starters relationships for the data (black) or simulations with parameters obtained from the model fit of the data (red), for one iteration of the fit. (B) Plot of experimental input fraction ratio between high and low starters vs relative p across areas obtained from fitting values.
Table 8.
Parameters obtained from fitting the probabilistic model to experimental data (from Fig 6).
Table 9.
Number of inputs from 2 brains using 3D detection or for simulated 2D detection from consecutive 50 μm wide slices, keeping every 4th slice, assuming cell radii of 10 μm.
Table 10.
Bounds per area for fitting the probabilistic model to experimental data (from Fig 6).