Optimisation of an exemplar oculomotor model using multi-objective genetic algorithms executed on a GPU-CPU...

Area of Science:

• Computational neuroscience
• Ophthalmology
• Biophysics

Background:

• Computational models are vital for understanding normal and abnormal eye movements.
• This study focuses on a neurobiological model of fast eye movements (saccades) and infantile nystagmus.

Purpose of the Study:

• To optimize an existing neurobiological model of saccades and infantile nystagmus using experimental data.
• To investigate the model's predictive capacity and identify potential modifications.

Main Methods:

• Utilized a multi-objective genetic algorithm for parameter optimization.
• Implemented a master-slave parallelization strategy distributing computations across a GPU and CPU.
• Performed direct fitting of the model to experimental saccade and nystagmus recordings.

Main Results:

• The model accurately simulated normal saccades, though a single parameter set could not fit all amplitudes.
• Identified parameter regimes for simulating canonical infantile nystagmus waveforms with high accuracy.
• Achieved a ~20x speedup in model integration using GPU parallelization compared to CPU.

Conclusions:

• Quantified the model's predictive power and suggested modifications for expanded behavioral simulation.
• Optimal parameter distributions support the hypothesis that saccadic braking signals cause infantile nystagmus.
• The developed parallelization method is adaptable for optimizing other complex computational biology models.