A SpineML Model of saccadic eye movements

This is the James et al. oculomotor model, which builds on many years of work carried out in Kevin Gurney's group at the ABRG at The University of Sheffield. The main paper for this model is "Integrating brain and biomechanical models - a new paradigm for understanding neuro-muscular control". The paper can be found at papers/integration/omm_integ.pdf. The published (open-access) version is at https://www.frontiersin.org/articles/10.3389/fnins.2018.00039/full

It builds on work which has been carried out by the following individuals: Jon Chambers, Alex Cope, Alex Blenkinsop, Seb James, Jen Lewis and indeed, Kevin Gurney. The biomechanics work - saccsim, our virtual eye - has been provided by Kostas Moustakas' group at The University of Patras: Chris Papapavlou and Dimitris Stanev were involved.

To run this model fully, and reproduce the results, you need the saccsim biomechanical eye, BRAHMS and SpineML_2_BRAHMS. To view the models, you need SpineCreator.

To install SpineCreator, BRAHMS, SpineML_2_BRAHMS (and also SpineML_PreFlight), please refer to the instructions at http://spineml.github.io/spinecreator/sourcelin/ Note that these are build instructions for Linux, and they explain how to build the following programs from github:

https://github.com/BRAHMS-SystemML/brahms

https://github.com/SpineML/SpineML_PreFlight

https://github.com/SpineML/SpineML_2_BRAHMS

https://github.com/SpineML/SpineCreator

Each of these repositories have a tag 'James_omm_integ' which tag the state of the repositories when the paper was submitted. In principle, the latest version of each should be able to reproduce the results of the paper, but to be sure, it is probably a good idea to checkout the James_omm_integ branches whilst building.

The saccsim eye model code as used for the paper results is available within this repository in the subdirectory c++/saccsim. Instructions for building the saccsim BRAHMS component is in c++/saccsim/README.md

To build our worldDataMaker component, and some tests of the c++ classes that are used in that code, do the following:

cd c++
mkdir build
cd build
cmake ..
make -j4

Now you'll have some BRAHMS components that you need to copy into your 'BRAHMS Namespace' which, if you've compiled brahms according to the instructions mentioned above, will be at:

/usr/local/var/SystemML/Namespace

Copy in the skeleton BRAHMS NoTremor namespace:

cp -Ra c++/brahms_namespace/NoTremor /usr/local/var/SystemML/Namespace/dev/

Now you can copy your newly compiled components into the NoTremor namespace:

worldDataMaker:

cp c++/build/worldDataMaker.so \
   /usr/local/var/SystemML/Namespace/dev/NoTremor/worldDataMaker/brahms/0/component.so

Centroiding components:

cp c++/build/centroid.so \
   /usr/local/var/SystemML/Namespace/dev/NoTremor/centroid/brahms/0/component.so
cp c++/build/powercentroid.so \
   /usr/local/var/SystemML/Namespace/dev/NoTremor/powercentroid/brahms/0/component.so
cp c++/build/multicentroid.so \
   /usr/local/var/SystemML/Namespace/dev/NoTremor/multicentroid/brahms/0/component.so

Saccsim:

cp c++/saccsim/src/build/BRAHMS/component.so \
   /usr/local/var/SystemML/Namespace/dev/NoTremor/saccsim/brahms/0/component.so

If I didn't miss anything out of these instructions, you should now be able to run the models (from the experiments interface of SpineCreator or from the SpineML_2_BRAHMS command line program 'convert_script_s2b'), which are:

ModelN models

The 'ModelN' models were those which were used to develop a training procedure for the SC_deep to LLBN weight maps. I also used some of these to investigate a widening projection field within the brain model, before abandoning this idea.

Model1: This is the model with no centroiding component, and no widening Gaussian projective fields. It's the naive model which demonstrates the issue whereby the retinotopic projection affects the size of the hill of activity in SC_deep (the experimental observation is that the size of this hill of activity is invariant wrt its position).

Model2: This is the model with the centroiding component between SC_deep and SC_avg and is the model used to compute the weight maps from SC_avg to the SBG

Model3: This is the model with a non-centroided connection from SCdeep to the SBG. It has a widening Gaussian projection from FEF_add_noise to FEF and from SC_sup to SC_deep. I've tried to optimise this model to make the best, most linear vertical movements (Rotations about X).

Model4: Like Model3, but parameters modified to give this the most linear horizontal movements (rotations about Y).

Model5: A copy of 4, which I will tweak to try to get vertical rotations to work reasonably well in conjunction with horizontal rotations.

TModelN models

The TModel0-2 models were a first set of models employing the theoretical weight mapping from Ottes et al. and Tabareau et al.

TModel0: A model with the theoretical SG->SBG transfer function as in Tabareau et al. Legacy.

TModel1: Like TModel0, but with a modification to improve the oblique saccades. Legacy.

TModel2: To be like TModel0 with WideningGaussian from SC_deep to SC_deep2 and a demonstrations that this resolves problems in TModel0. Legacy.

From the perspective of the paper, TModel3-5 are the only ones that matter.

TModel3: Back to the naive model - Model1 - but with the exponential/sine wave weight maps predicted by the Ottes mapping and Tabareau et al. 2007.

TModel4: To TModel3 is added a widening Gaussian projection field from SC_deep to a new layer, SC_deep2.

TModel5: This adds a mechanism to reset the TN firing in the SBG, which was missing in the first iteration of the SBG.

Experiment scripts

The 'experiments' are carried out using scripts in the batch_scripts subdirectory. In there, there's a set of expt_* dirs. Typically, these contain a shell script to run the simulations, and some octave/matlab (octave preferred, you may need to tweak if you use matlab) scripts to plot the results for a quick view. There are Veusz files scattered about that were used to generate the figures in the paper.

Within batch_scripts there are also two directories containing octave code for running the sims; analysis and include.

Note that some of the sim scripts were set up to run on my laptop/PC but most were set up to run on Sheffield's HPC system Iceberg, which uses Son of Grid Engine to manage jobs.

lab_book sub-directory

My labbook is found in the lab_book subdirectory.

matlab sub-directory

This contains some old code that was used to generate matlab movies

movies sub-directory

Contains scripts to generate movies from a sim run.

papers sub-directory

Contains two papers that I wrote about this project; a mini-conference paper and the main integration paper.

Providence

This repository was extracted from Oculomotor branch of our private abrg_local repository on 3rd May 2017, with a view to making it public with the paper. Earlier history on model/code changes is held there.