Inferring and validating mechanistic models of neural microcircuits based on spike-train data

Abstract

The interpretation of neuronal spike train recordings often relies on abstract statistical models that allow for principled parameter estimation and model selection but provide only limited insights into underlying microcircuits. In contrast, mechanistic models are useful to interpret microcircuit dynamics, but are rarely quantitatively matched to experimental data due to methodological challenges. Here we present analytical methods to efficiently fit spiking circuit models to single-trial spike trains. Using derived likelihood functions, we statistically infer the mean and variance of hidden inputs, neuronal adaptation properties and connectivity for coupled integrate-and-fire neurons. Comprehensive evaluations on synthetic data, validations using ground truth in-vitro and in-vivo recordings, and comparisons with existing techniques demonstrate that parameter estimation is very accurate and efficient, even for highly subsampled networks. Our methods bridge statistical, data-driven and theoretical, model-based neurosciences at the level of spiking circuits, for the purpose of a quantitative, mechanistic interpretation of recorded neuronal population activity.

Training deep neural density estimators to identify mechanistic models of neural dynamics

“…For some mechanistic models in neuroscience (e.g. for integrate-and-fire neurons), likelihoods can be computed via stochastic numerical approximations ( Chen, 2003 ; Huys and Paninski, 2009 ; Meliza et al, 2014 ) or model-specific analytical approaches ( Huys et al, 2006 ; Hertäg et al, 2012 ; Pozzorini et al, 2015 ; Ladenbauer et al, 2018 ; René et al, 2020 ).…”

The role of mental simulation in primate physical inference abilities

“…The strongest and most recent evidence comes from studies that show a certain degree of consistency between human behavior and the behavior of high-level computer programs running simulations (Battaglia, Hamrick, and Tenenbaum 2013;Ullman et al 2017) . Whether such high-level computer programs are suitable abstractions for how neural systems compute is debatable (Ladenbauer et al 2019) . Second, while there has been significant progress in creating model-based neural network agents (Goodfellow et al 2014;Higgins et al 2016;Kulkarni et al 2015) , exerting flexible control over such neural models has proven challenging (Nalisnick et al 2018) .…”

Uncovering Network Architecture Using an Exact Statistical Input-Output Relation of a Neuron Model

“…Although there have been attempts to include common inputs from unobserved neurons into the GLM framework by treating them as hidden variables (Kulkarni and Paninski, 2007; Vidne et al, 2012), variations in the structure of hidden common inputs are limited. In addition, these statistical models are not directly constrained by physiologically plausible membrane dynamics and spiking threshold while the LIF neuron model is (Ladenbauer et al, 2019). Here, given knowledge about the balanced network, we introduce hidden inputs as background noise and additionally consider various architectures of arbitrarily strong hidden common inputs as shared signals.…”