The Mecp2 Project

We (Dr. Susanna Mierau, Riccardo Conci, and Timothy Sit (I), from the Neuronal Oscillations Group, with the help of Dr. Guillaume Hennequin from the Computational and Biological Learning Lab, and Dr. Manfred Kitzbichler from the Behavioural and Clinical Neuroscience Institute, University of Cambridge, UK) are investigating the network properties that arises during neurodevelopment, and how they are disrupted in neurodevelopmental disorders. We are currently using a mice model of autism (MECP2 deficient mice, hence the title of the project) as a starting point for investigating differences in network properties in these neurons. We are particularly interesting in the topology of the functional connectivity network (Riccardo's focus) and how the dynamics of this network is controlled (my focus).

The analysis here is divided into those based on R (mainly statistical analysis, but I did do some early spike analysis there) and those based on matlab, the main script to run for R is main.Rmd in the R_analysis folderl. I am still working on a main script for matlab analysis, but all the required functions can be found inside the Matlab_Analysis folder. The R script will ask you to select a folder containing the h5 and/or .mat files you want to analyse. I have also performed some preliminary statistical analysis which you can find in the R notebook inside the Statistical_Analysis folder.

Thesis / Project Report

Now available here!

Poster

Poster presented at the Cambridge Neuroscience Symposium - Mental Plasticity & Development (March 2018)

R notebook

Rendered version available here Source code available in R_analysis folder.

Workflow for R analysis

Requirements

• computer running unix (Mac or Linux)
• Matlab (for converting raw .mcd files to .h5 files with spikes only)
• A suitable C compiler in Matlab (2017 matlab doens't seem to work on my computer for some reason, but it may be just me. It worked when I used an older versin of matlab/C compiler.)
• R and RStudio (for spike analysis)
• Stephen Eglen code and packages (need to give more details on this)

Overview

---

• we start with the raw .mcd files
• we run Prez's code to get only spikes (assuming you are doing spike analysis, if you want to do oscillation stuff, see workflow for matlab analysis. Alternatively, you can write/find code for oscillation analysis in R.)
• currenlty we don't do spike sorting, if you want to do that, look into WaveClus
• we then use Eglen's code (and some code written by myself), for spike anlaysis

---

The steps from the recorded data to feature matrix X should be the following:

• record and get .mcd files (usually 12 minutes long, about 2GB per file)
• use WaveClus (Prez's code) for spike detection, this will create h5 files that have the detected spikes (matlab, unix only)
• Run Eglen's code in to get features (R)
• may need to look into something for spike sorting (not sure if Eglen's code require spikes to be sorted or not)

alternative way is to use MCRack (windows only), seems to suport most of the data processing mentioned above but not sure why Eglen didn't use them.

Running Prez's code for spike detection

• raw data = .mcd file
• Prez wrote code to find spike times, this is converted to HDF5 format (the code only works on Unix systems)
  • https://github.com/przemyslawj/mcdtools
  • Klusta mentioned in dependencies but not needed, WaveClus is sufficient
• saved data fields described in the paper A data repository and analysis framework for spontaneous neural activity recordings in developing retina
  • names should be intuitive enough so no need to study in detail

The instructions for running Prez's code is in the following link: https://github.com/przemyslawj/mcdtools But I will reiterate them here just in case.

Steps:

• Download waveclus in your linux operating system (you know what I mean, don't talk to me about GNU/Linux)

wget https://github.com/csn-le/wave_clus/archive/testing.zip
unzip wave_clus-testing.zip

• put all the .mcd files you want to convert to h5 into one folder
• make sure the file names are in the format described below
• open matlab
• change directory to your "project folder" (for me, I created a folder called "The Mecp2 Project" and put everything relating to the project inside).
  • Put the mcdtools-master folder into your project folder
  • Put the folder containing your .mcd files in your project folder
  • ie. your project folder should now contain the folders mcdtools-master and whatever name of the folder containing your .mcd files
• add everything in the the mcdtools-master to path, (including subfolders)
• add your folder containing .mcd files to path

.mcd file format for Prez's code

Note that the file format has to be this:

{TYPE}_{DATE}-{MEA_ID}_DIV{DAY_IN_VITRO}.mcd

Where:

• TYPE is the condition distinguishing the recordings, e.g. KO or WT,
• DATE is the date of culture initiation in DD_MM_YY format, e.g. 03_05_17
• MEA_ID is in [1-9][A-Z] format where the digit corresponds to a pup ordinal and the letter is the ordinal of the dish initiated from that pup, e.g. 4C
• DAY_IN_VITRO - DIV of the culture during the recording. Note that two digits are expected, ie. if the culture is two days in vitro, then write 02 instead of 2.

you may find that some previous file format is in the form:

tc042_d24.mcd

Overview of workflow

Results in pictures

A quick summary of the progress so far, in figures.

We start with a set of 60 voltage recordings from each micro-electrode array:

Overview of raw data and detected spikes

Overview of spike activity in electrodes

Spike distribution

Network weighted by covariance

Effective Rank

Comparing spike detection algorithms

Algorithms we are testing:

• simple threshold-based methods
• nonlinear energy operator followed by threshold
• continuous wavelet transform
• (in progress) Gaussian Process Spike Detection

Note that Prez's algorithm is based on Wave_Clus.

A close look at what's happening

Tuning thresholds

Network Analysis

The schemaball connectivity graph is plotted using Paul Kassebaum's circularGraph function, which can be found here

Dynamic network is made based on the work by Sizemore and Bassett which can be found here.

Work in progrss

• better networks
• network anlaysis
• auto and cross-correlation
• embed interactive plotly histogram
• Gaussian Process Classification, Feature Selection

Resources

Python library for electrophysiological signals analysis http://elephant.readthedocs.io/en/0.2.0/index.html

Spike train analysis with R https://cran.r-project.org/web/packages/STAR/STAR.pdf