NeuralComp

In neuroscience research, a theory known as multiplexing posits that when presented with multiple stimuli together, individual neurons can switch over time between encoding each member of the stimulus ensemble, causing a fluctuating pattern of firing rates. NeuralComp is an R package for testing and analyzing rate fluctuations of multiplexing neurons under a Bayesian paradigm.

Corresponding Paper

• Modeling Neural Switching via Drift-Diffusion Models

Data format

To gain insight into multiplexing, scientists often collect triplets of data consisting of: spike trains recorded under an $A$ stimulus, spike trains recorded under a $B$ stimulus, and spike trains recorded under both the $A$ and $B$ stimuli (we will refer to this as the $AB$ condition). An example of a triplet from Caruso et al. can be seen below. Given a triplet, we wish to infer whether the neuron utilizes multiplexing to encode both stimuli and, if so, at what time-scale does the neuron switch between encoding the two stimuli.

Practical Consideration: We often necessitate at least 5 trials from each condition in the triplet.

Models

In the corresponding paper, we propose a mechanistic statistical model for multiplexing. The model uses the integrate-and-fire framework as the basis, which means that we assume that the spikes occur as a result of a latent membrane voltage process hitting a threshold. In this work, the latent membrane voltage process is assumed to be a drift-diffusion process. By assuming a perfect-integrator integrate-and-fire model, we have that the hitting times of the drift-diffusion process are inverse Gaussian distributed. The relationship between the latent drift-diffusion process and the spikes can be seen below. We will model the $A$ and $B$ condition spike trains using these types of inverse Gaussian point processes, and will posit that multiplexing occurs due to competition between the $A$ condition drift-diffusion process and the $B$ drift-diffusion process. To control the overall firing rate and rate of switching, we will assume that there is some inhibition (or penalty for switching) in the form of a time delay ($\delta$) on one of the processes. The effect of $\delta$ can be seen in the figures below. This type of model results in a poential neural motif that could lead to this type of behavior (as seen in subfigure B).

Alternative Model: We propose a alternative model that characterizes alternative encoding schemes (normalization, winner-take-all, ect.) with some level of generality. This model assumes that the $AB$ condition spike trains can be modeled using another inverse Gaussian point process with parameters not necessarily relating to the $A$ process or the $B$ process.

Conceptual Idea

Given our multiplexing model and alternative model, we use WAIC to determine whether the data supported the occurance of multiplexing in the $AB$ trials. The general schematic of the analysis can be seen in the figure below.

Associated Repositories

1. Simulation Studies
2. Case Study

Related Multiplexing Papers

Statistical Modeling

1. Spike Count Analysis for MultiPlexing Inference (SCAMPI)
2. Analyzing second order stochasticity of neural spiking under stimuli-bundle exposure

Neuroscience Papers

1. Signal switching may enhance processing power of the brain
2. Single neurons may encode simultaneous stimuli by switching between activity patterns
3. Coordinated multiplexing of information about separate objects in visual cortex
4. Multiple objects evoke fluctuating responses in several regions of the visual pathway
5. Sensitivity and specificity of a Bayesian single trial analysis for time varying neural signals