Epoch dependent linear dynamics system fit
This code of epoch-dependent linear dynamics system (EDLDS) fit, was originally developed and used to uncover the latent-space dynamics from the simultaneous recordings of frontal neural dynamics in Wei et al. (2019).
- Code --- LDSI code version 3.0 (Matlab); Potentially changed to Python in later version.
- TestExamples-ArtificialData --- A list of files used for tests of simulated neural system
- TestExamples-NeuralData -- A list of files used for fitting an exemplary simultaneous neural recording session
#TestExamples-ArtificialData Folder
- SimulateLDS.m --- A code for simulated neural activity with single stage
- SimulateLDS_MultiStage.m --- A code for simulated neural activity with multiple stage
- stochastic.m --- A code from SSID (subspace identification) for estimates of parameters
-
Test_State_Variable.m --- Test for goodness of fits for a one-dimensional latent variable system. For this case, one can in fact exam whether the fitted latent system is the similar to the original one. In plot, the black line stands for the original latent dynamics, and the red line stands for the fitted latent dynamics. In the default setting, there are 4 stages (and thus the breaking points are those between stages).
-
Test_Arti_Neural_Data.m --- Test for goodness of fit for a single stage one-dimensional latent variable system, where spike count is either generated by a Poisson (Part 1 of the code) or Log-normal (Part 2 of the code) process. Specifically,
Poisson spike count generation
Spike counts: Y(t) = Poiss(exp(Z(t)))
Spike rate: Z(t) = Cx(t)
Latent dynamics: x(t) = Ax(t-1) + w(t)
Log-normal spike count generation
Spike counts: Y(t) = Round(exp(Z(t)))
Spike rate: Z(t) = Cx(t) + v(t)
Latent dynamics: x(t) = Ax(t-1) + w(t)
Here is an exemplary comparison between original and fitted latent neural dynamics using Log-normal after running this code.
-
Test_Arti_Neural_Data_Multiple_Latent_Variables.m --- The same as Test_Arti_Neural_Data.m, expect it generates the data for multiple stages.
-
Test_Arti_Neural_Data_Percentage_Explained.m --- The same as Test_Arti_Neural_Data.m, expect it generates the data from multi-dimensional latent variables (xDim > 2) and one stage. Plots includes comparison of explained variance, BIC and AIC as fitted by different numbers of latent dimensions. Also the comparison of Y (grey lines),
$Y_{exp}$ (red lines) and$Y_{est}$ (blue lines).
Simulated neural data: Y(t) = Cx(t) + v(t)
Latent dynamics: x(t) = Ax(t-1) + w(t)
Expectation of simulated neural data:
$Y_{exp}$ (t) = Cx(t)Estimation of neural data:
$Y_{est}(t) = \bar{C} \bar{x}(t)$
Test_Nuo_data_analysis_LDS.m --- Example fits using data ANM219033_20131116.mat (which is already preprocessed).
- original data
- fitted data using EDLDS model
MIT
Wei et al. (2019): An orderly single-trial organization of population dynamics in premotor cortex predicts behavioral variability. Nature Communications 10, 216 (2019).