Synthesis methodology

This page presents the scientific procedure used to synthesize the cells.

Overview

Production workflow

Basically, the workflow used in BBP to generate the cell morphologies is the following:

• collect biological data,

• calibrate the models on these data,

• synthesized the cells according to the calibration parameters.

This package aims at running the calibration process, then calling each part of the production workflow on a small use case and computing some validation metrics. Once the calibration parameters are correct, the actual production workflow can be run using them.

Internal workflow

The main workflow that should be used is: tasks.workflows.ValidateSynthesis. This workflow runs all the required tasks to calibrate and validate the synthesis models. It is divided in several subtasks (more details on these tasks are given in tasks):

Calibration parameters

The calibration step should create the two parameter files used by placement_algorithm.app.synthesize_morphologies.Master:

• tmd_parameters.json which contains the model parameters for each mtype.

• tmd_distributions.json which contains the distribution properties of each mtype.

Details on the content of these files can be found here: <URL> (does not exist yet)

Synthesis in atlas

When cells are synthesized inside an atlas, their shapes must be adapted according to their positions in order to fit in this atlas. Currently, the shapes are just rescaled in order to fit in a defined interval. This interval depends on the mtype and on the cell position because the depths of the atlas layers also depend on this position. The information on how each mtype can be rescaled are inserted in the tmd_parameters.json file by the task tasks.synthesis.AddScalingRulesToParameters.

For a given mtype, the parameters specific to the scaling process are contained in a context_constraints key. The keys in this object should be neurity types (one of [apical, basal, axon]) and should contain the interval in which it must fit. This interval is defined relatively to the atlas layers, so its two boundaries should contain a layer entry, which should be an integer in [1, 6], and an optional fraction entry, which should be a float in [0, 1] (0 for the bottom boundary of the layer and 1 for the top boundary).

For apical rescaling, another optional extent_to_target entry should be added. This entry contains a target layer entry and an optional fraction entry as described before. But it also contains the fit parameters of the path distance of the cell as a function of its extent along its principal direction. This is needed because the stopping criterion used in the TNS model only takes the path distance into account. This fit is linear and is thus described by a slope and a intercept entries.

Here is an example of such context_constraints entry:

{
    "<mtype>": {
        "context_constraints": {
            "apical": {
                "extent_to_target": {
                    "fraction": 0.8,
                    "intercept": 0,
                    "layer": 1,
                    "slope": 1.5
                },
                "hard_limit_max": {
                    "fraction": 1,
                    "layer": 1
                },
                "hard_limit_min": {
                    "fraction": 0.1,
                    "layer": 1
                }
            },
            "basal": {
                "hard_limit_max": {
                    "fraction": 0.5,
                    "layer": 1
                }
            }
        }
    }
}

More details on the models can be found here:

• NeuroTS

• region-grower

• placement-algorithm