Aplysia Feeding Kinetic Model

Code repository: https://github.com/CWRUChielLab/AplysiaFeedingKineticModel

This repository contains code for a detailed biomechanical model of the feeding apparatus of the sea slug Aplysia californica. See Snyder (2005) for a detailed "Guidebook to the Model" defining equations and parameters:

Snyder, V. A. (2005). Analysis of the biomechanics and neural control of two kinetic models of the buccal mass of Aplysia [M.S. thesis, Case Western Reserve University]. http://search.proquest.com/docview/305390183/abstract/B303E6FBC4924763PQ/1

The model is instantiated in C++, and visualization scripts are written in Python.

Usage

1. Navigate to the project directory:

cd AplysiaFeedingKineticModel

2. The Makefile provides convenient recipes for compiling the code, batch-running the model, and validating results. To simply compile the source code, run the following:

make

Optionally, use make debug to compile the code with extra print commands for debugging. The make clean command can be used to clean up your working directory by removing compile artifacts and model outputs.

3. Run the model executable:

./model <behavior>

where <behavior> is one of several valid behavior types that can be simulated, such as Bite. For a full list of possible behavior types, run ./model without arguments. Each type differs in the activation timing of muscles, resulting in different movements.

Available behavior types:

• Bite: A bite behavior (failure to grasp food)
• SwallowA: A type-A swallow (smaller amplitude)
• SwallowB: A type-B swallow (larger amplitude with dorsal rotation of the held food)
• SwallowPerturbed: A modified type-B swallow with a brief external force applied to the food
• RejectionA: A type-A rejection (smaller amplitude)
• RejectionB: A type-B rejection (larger amplitude with ventral rotation of the held object)
• IzhikevichBite: Similar to Bite, but with a layer of integrate-and-fire motor neurons driving the muscles at defined frequencies
• IzhikevichSwallowA: Similar to SwallowA, but with a layer of integrate-and-fire motor neurons driving the muscles at defined frequencies
• IzhikevichSwallowB: Similar to SwallowB, but with a layer of integrate-and-fire motor neurons driving the muscles at defined frequencies
• IzhikevichSwallowPerturbed: Similar to SwallowPerturbed, but with a layer of integrate-and-fire motor neurons driving the muscles at defined frequencies
• IzhikevichRejectionA: Similar to RejectionA, but with a layer of integrate-and-fire motor neurons driving the muscles at defined frequencies
• IzhikevichRejectionB: Similar to RejectionB, but with a layer of integrate-and-fire motor neurons driving the muscles at defined frequencies
• IzExampleSwallow: A more advanced model of the neuronal inputs to the muscles, where many identified motor neurons are represented
• IzSwallowBmoreaccurate: Similar to IzExampleSwallow (?)
• Dynamic: A version that uses a user-customizable input file (CSV format) to specify the timing and magnitude of inputs to the muscles
• NeuromechanicalInput: Similar to Dynamic, but expects a different set of inputs created by another model of the motor neurons with interneurons coordinating them (https://github.com/CWRUChielLab/AplysiaNet)

Integrate-and-fire neurons were implemented following Izhikevich (2003):

Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks, 14(6), 1569–1572. https://doi.org/10.1109/TNN.2003.820440

Visualizing Results

When the model runs, it generates output files in CSV format containing periodic snapshots of the values of various quantities, including positions, angles, and levels of muscle activation. Python scripts can generate plots and animations from these output files:

• newanimation.py: generates movies of the motions of the buccal mass, with time plots of the muscle inputs
• PlotVariablesSlugOutput.py: generates time plots of the kinematic variables
• makerasterplot.py and PlotVariablesIzhikevich.py: generate figures of the activity of the integrate-and-fire motor neurons, which are not included in all behavior types

The make animations and make figures commands can be used to batch-run the model on a subset of the behavior types, renaming and saving the resulting movies and figures to dedicated directories.

The Python scripts require a handful of packages that can be installed with pip:

pip install numpy pandas matplotlib tqdm

Validation

The outputs of the model when run on a subset of the behavior types are saved in the Check directory. The make check command can be used to systematically compare your version's outputs to these "gold standards". This is useful for catching differences in implementation of libraries across platforms, or other discrepancies that might cause the model to run differently across systems. Deviations should be investigated as they may lead to bugs.

Of course, if the code is modified intentionally in such a way that the outputs change (e.g., a new column is added to the output files), the "gold standard" against which make check runs will need to be updated. The make bless command will systematically run the model on the subset of behavior types and copy the outputs to the Check directory. These changed outputs, once verified to be correct, should be committed to the repository so that results can be validated against them in the future (they will become the new "blessed" gold standard of what outputs the model should produce).