Supercomputers and Reverse Engineering of Motoneuron Firing 1 Patterns

In this study, we achieved a major step forward in the analysis of firing patterns of populations of motoneurons, through remarkably extensive parameter searches enabled by massively-parallel computation on supercomputers. The ability to implement these extensive parameter searches seem ideally matched to understanding the many parameters that define the inputs to neuron populations that generate these patterns. Therefore, we investigated the feasibility of using supercomputer-based models of spinal motoneurons as a basis for reverse engineering (RE) their firing patterns to identify the organization of their inputs, which we defined as the amplitudes and patterns of excitation, inhibition, and neuromodulation. This study combines two advances: 1) highly-realistic motoneuron models based on extensive in situ voltage and current clamp studies focused on neuromodulatory actions, and 2) implementation of these models using the Laboratory Computing Resource Center at Argonne National Laboratory to carry thousands (soon millions) of simulations simultaneously. The goal for computing and performing RE on over 300,000 combinations of excitatory, inhibitory, and neuromodulatory inputs was twofold: 1) to estimate the synaptic input to the motoneuron pool and 2) to generate training data for identifying the excitatory, inhibitory, and neuromodulatory inputs based on output firing patterns. As with other neural systems, any given motoneuron firing pattern could potentially be non-unique with respect to the excitatory, inhibitory, and neuromodulatory input combination (many input combinations produce similar outputs). However, our initial results show that the neuromodulatory input makes the motoneuron input-output properties so nonlinear that the effective RE solution space is restricted. The RE approach we demonstrate in this work is successful in generating estimates of the actual simulated patterns of excitation, inhibition, and neuromodulation with variances accounted for ranging from 75% to 90%. It was striking that the nonlinearities induced in firing patterns by the neuromodulation inputs did not impede RE, but instead generated distinctive features in firing patterns that aided RE. These simulations demonstrate the potential of this form of RE analysis. It is likely that the ever-increasing power of supercomputers will allow increasingly accurate RE of neuron inputs from their firing patterns from many neural systems.