Monte Carlo validation of spike-timing-dependent plasticity identification from spiking activity.

Abstract

As increasingly complex non-stationary models become possible to be identified from neural data, rigorous validation approaches must be developed to rule out overfitting and the potential to identify features by chance. Specifically, identification of spike-timing-dependent plasticity (STDP) from recorded spontaneous in vivo spike timing is a potentially powerful tool to quantify activity-dependent plasticity. In previous work, we presented a methodology to perform this STDP identification from spike timing alone and successfully identified a generative model. Validation was straightforward with the generative model because the underlying model was known, but becomes challenging when applied to experimental data. Here, we introduce a set of null hypothesis tests that can be performed with Monte Carlo (MC) simulations of null models to rule out cases of overfitting with experimental data. We demonstrate the identification of these null models and null hypothesis testing on a generative model in two test cases, one with and one without overfitting. Importantly, we show that it is possible to distinguish an identified STDP rule from a null case where there are similar weight fluctuations which are activity-independent. With the development of the null hypothesis tests described here, STDP identification can be effectively applied to experimental data recordings.