This repository contains code for the high-level analysis of the functional connectivity data in Randi et al. (submitted). It contains several scripts to generate figures, that rely on the 3 main classes Funatlas, Fconn, and ExponentialConvolution, as well as the Python modules wormdatamodel and wormbrain. The figures of the paper are reproduced using the commands found in the scripts/fconnectivity/figures/paper/ folder.
Funatlas is the functional atlas class that aggregates functional connectivity data from multiple datasets from different animals. The class can be instantiated via the regular constructor or from the class method from_datasets(), which will load the data from several datasets and compile the results based on neural identities. Funatlas can be instantiated to maintain the "exact" neural identities, or to merge neurons into classes (i.e. to approximate neuron identities):
merge_bilateral=True will merge results for, e.g., AVAL and AVAR into the class AVA_,
merge_dorsoventral=True will merge RMED and RMEV into RME_, while
merge_numbered=True will merge VB3, VB4, ... into VB. These options can be combined to merge, for example, SMBVL, SMBVR, SMBDL, and SMBDR into SMB__ with
Important methods regarding neuron identities are
Funatlas.ids_to_ai() converts neuron identities to atlas-indices (ai), given the neuron-class-merging options with which the object has been created.
Funatlas.i_to_ai() converts dataset-specific indices of neurons into atlas-indices. For example, in a given dataset, neuron ADAL could be neuron number 38, but in the atlas list of IDs, ADAL is number 0 (i.e. its atlas-index is 0).
The main goal of the Funatlas is to aggregate data from many datasets, for example obtaining all the responses of neuron AVER to stimulations of neuron AVDL. When Funatlas is created using
funa = Funatlas.from_datasets(...)
you can access the
pumpprobe.Fconn class for each recording ds (see below) as
wormdatamodel.Signal class as
funa.sig[ds], and the
wormbrain.Brains class as
funa.brains[ds]. You can also obtain the matrices occ1, occ2, and occ3 using the functions
occ1, occ2 = Funatlas.get_occurrence_matrix()
occ3 = Funatlas.get_observation_matrix()
occ1[i,j] is the number of times neuron i responded to stimulations of neuron j,
occ3[i,j] is the total number of times one could have observed a response of i to stimulation of j (regardless of whether there was a response or not).
occ2[i,j] is a list of dictionaries that contain the relevant information to retrieve the neural activity traces of the responses of i to stimulations of j.
Many other methods return aggregated results from the datasets, like
Funatlas.get_signal_correlations(), in addition to many more, including utilities functions.
Finally, Funatlas also has methods to load and use the C. elegans anatomical connectome (from White et al. 1986 and Witvliet et al. 2020), the known extrasynaptic connectome (from Bentley et al. 2016), and gene expression data from CeNGEN (Taylor et al. 2021).
The Fconn class contains functional connectivity data coming from an individual recording. In addition to the automatic response detector, Fconn contains the fits of the kernels describing functional connectivity. These are the kernels kij(t) that give the activity of the downstream neuron i when convolved with the activity of the stimulated neuron j. Kernels are represented symbolycally by the ExponentialConvolution class (see below). Fconn both stores the results of the fits in
Fconn.fit_params, and has methods to fit the kernels (
The class ExponentialConvolution symbolically represents sums of convolutions of decaying-exponential response functions. In addition to the symbolic calculation of the convolutions (
ExponentialConvolution.convolve_exp()) and its evaluation in the time-domain (
ExponentialConvolution.eval()) and in the frequency-domain (
ExponentialConvolution.eval_ft()), the class also contains methods that return the effective rise time and decay time of the kernel (
ExponentialConvolution.drop_saturation_branches() removes from the symbolic expression the terms associated with saturation of signal transmission.
ExponentialConvolution contains several methods to simplify the symbolic expression via clustering of the timescales of the decaying exponentials and matching similar timescales across different instances of the class.