Scaling of gene transcriptional gradients with brain size across mouse development

This repository contains Matlab and python code for reproducing analyses presented in the open-access journal article:

📗 H.Y.G. Lau, A. Fornito, B.D. Fulcher. Scaling of gene transcriptional gradients with brain size across mouse development, NeuroImage 224: 117395 (2021).

Additional code to reproduce simulations of a simple physical model is in this GitHub repo.

Data was taken from the Allen Institute's Developing Mouse Brain Atlas using python scripts that call the AllenSDK (in DataRendering).

A description of the raw data retrieval via the Allen SDK is at the bottom of this document.

Processed data is available from this Zenodo data repository, and can be downloaded to reproduce the results presented in our paper (using the functions outlined below). (FYI: a previous related dataset was uploaded to this figshare repository).

Results

Before running analyses, add all repository paths using startup.

Plotting voxel expression data

Clustered gene-expression matrices

You can get the clustered gene-expression plots as:

PlotAllExpressionMatrices

Or for each individual one, as, e.g., PlotExpressionMatrix('E11pt5'). Saves to Outs/ExpressionMatrix_E11pt5.png:

Three-dimensional voxel principal components

PlotAll3dSpatial

And individual time points as VisualizeSpatialExpression.

For example, taking E18.5 (without distinguishing fore/mid/hindbrain) and subsampling to 5000 voxels we can see the two gene-expression PCs for the left hemisphere in three-dimensional space:

VisualizeSpatialExpression('E18pt5','','turboOne',5000)

CGE(d) Curves

makeCGECurves()

Yields Fig. 3:

Dependence on anatomical subdivisions:

We can also modify the data used for these computations, including subsets of voxels and/or genes.

For example, to compute the same curves but only using forebrain voxels and neuron-enriched genes:

params = GiveMeDefaultParams();
params.thisBrainDiv = 'forebrain';
params.thisCellType = 'neuron';
makeCGECurves(params)

Contributions from voxels within or between major anatomical subdivisions

We can also investigate the contribution of pairs of voxels within and between major anatomical divisions in one go:

params = GiveMeDefaultParams();
WithinBetween(params,true,true,true);

You can get clearer individual figures by saving each individual time point to a separate figure as: WithinBetween(params,true,false,true);.

Scaling relationships

First compute the exponential fitting results using ComputeFittingResults.

Then you can visualize the results as:

makeConstantPlot();

Produces the plots of parameter scaling:

Human prediction

You can use mouse data to make predictions about the spatial transcriptional correlation length in human:

PredictHuman();

The dependence of lambda estimates on voxel subsampling can be assessed as:

VoxelSamplingDependence()

Data

Raw data retrieval

Raw data was retrieved via the API made available using the Allen SDK. Code is available in this repository:

1. Download Allen API package by pip install allensdk
2. Run download_devmouse_unionizes.py to retrieve gene expression data at structure level
3. Run structures.py to download structure information
4. Run getBrainDivision.py to download major brain division info (forebrain, midbrain and hindbrain) and ID of their descendants; data saved in structure_F.csv, structure_M.csv,structure_H.csv,structure_F_descendant_ID.csv,structure_M_descendant_ID.csv,structure_H_descendant_ID.csv.

Processing of raw data

The processed data files were obtained by running createData(false), which creates data files starting from energyGrids.mat variables from this work's associated Zenodo data repository.

(FYI, previously this figshare repository).

The full pipeline from raw data (downloaded via the Allen SDK) can be reproduced using createData(true).

You can also run createVariance to create the data of variance in decay constant against number of data points used (takes a long time, >24h).