Sorting of the Locust Olfactory Pathway Datasets from Zenodo

Table of Contents

1 Introduction

You will find here the codes necessary for reproducing the sorting of the extracellular recordings from the locust olfactory pathway available on zenodo ( as well as the sorting results (the spike trains).

For now the analysis is done with R but a C version is in preparation. The way the code is applied to the specific datasets can be followed with `html` files that were generated using Emacs Org mode, especially the Babel extension of Org mode (the developers of these wonderful tools are warmly thanked! Without their work none of what follows would have been possible–this does not mean that they should be held responsible for my mistakes–).

There is one directory per experiment on the dedicated GitHub repository. In the directory of each experiment you will find:

  • A file named Sorting_XXX_tetY.html, where XXX is the date (corresponding to the dates on the zenodo repository) and Y identifies the tetrode (there are four possible ones, A, B, C and D). This is the file you should look at first if you want a description of the sorting and if you want to judge its quality.
  • A file named, this is the source file from which the html version was generated.
  • A directory named locustXXX_spike_trains that contains the individual spike trains.
  • A directory named locustXXX_tetY_fig that contains the figures of the html file.

The spike trains in directory locustXXX_spike_trains are stored in ASCII format with one spike time (in seconds) per line. They are named locustXXX_StimID_tetY_uZ.txt, where XXX and Y are their previous meaning, StimID is a stimulation identifier (more precisely a group name in the HDF5 data file) and Z is the unit number. When several trials were recorded, like say 25 stimulation with citronellal, the successive trials will be found one after the other and time 0 is defined as the start of the acquisition of the first trial.

2 The code

There is a "reference manual" describing the R function specifically developed for spike sorting. A more detailed description of what these function do is also available.

3 The experiments

There are one or several html files (and org source files) per experiment–one if a single tetrode was recorded and more if two or more were recorded–. Each html file is an "sorting lab book", you will find there every command used to carry out the sorting together with comments and diagnostic plots. Each HDF5 group in the data file is analyzed trial per trial. The templates are allowed to drift to track the units (see the "reference manual" for info on how this is done). The diagnostic plots show for each unit in the model:

  • the successive templates,
  • the number of events attributed it (as well as the number of unclassified events),
  • the observed counting process,
  • the estimated ISI density,
  • the raster plot (when that makes sense, that is when odors were applied).

Feel free to critisize the procedure and its results. If you manage to improve it, be cool and let me know!

3.1 locust20000214

An experiment with a signle recorded tetrode and two well isolated neurons, see the description for details. In that experiment the inter trial interval was 30 s and only 10 s were recorded per trial. In the stored spike trains I have but the successive trials with a 10 s offest (corresponding to the acquisition time) and not with a 30 s (as described above, corresponding to the inter-trial interval). This makes the rater plot look nicer, nothing more.

3.2 locust20000421

An experiment were two tetrodes with reasonnably good signals were recorded: tetrode D1 with a single unit that can be followed through the 21 different stimulation and tetrode D2 with two units that can be followed. The data for this experiment are "unstable" because, I think, the odors were presented at too high a rate, once every 10 seconds.

3.3 locust20000613

An experiment were only tetrode D was giving good data. The analysis of this experiment is interesting because between two sets of trials (between stimulation with pure and 100 times diluted Cis-3-Hexen-1-ol), one waveform drifts so much that the old template does not recognize it anymore. This shows up in the diagnostic plots and the model has to be adjusted (by hand). Overall, 4 well isolated units are followed.

3.4 locust20000616

An experiment one can forget, tetrode C has 4 units, two of them are well isolated and the odor responses are not very stable.

3.5 locust20010214

This is probably the best experiment contained in the zenodo datasets collection. The details of the analysis of tetrode B are available. 10 units are identified, but only the first seven are well isolated. There is both spontaneous activity and responses to many different odors.

3.6 locust20010217

Only spontaneous activity. The details of the analysis for tetrode D are available. 10 units are identified but only 5 are good.

Author: Christophe Pouzat

Created: 2016-12-15 jeu. 16:59