BluePyOpt: Leveraging Open Source Software and Cloud Infrastructure to Optimise Model Parameters in Neuroscience

Abstract

At many scales in neuroscience, appropriate mathematical models take the form of complex dynamical systems. Parameterizing such models to conform to the multitude of available experimental constraints is a global non-linear optimisation problem with a complex fitness landscape, requiring numerical techniques to find suitable approximate solutions. Stochastic optimisation approaches, such as evolutionary algorithms, have been shown to be effective, but often the setting up of such optimisations and the choice of a specific search algorithm and its parameters is non-trivial, requiring domain-specific expertise. Here we describe BluePyOpt, a Python package targeted at the broad neuroscience community to simplify this task. BluePyOpt is an extensible framework for data-driven model parameter optimisation that wraps and standardizes several existing open-source tools. It simplifies the task of creating and sharing these optimisations, and the associated techniques and knowledge. This is achieved by abstracting the optimisation and evaluation tasks into various reusable and flexible discrete elements according to established best-practices. Further, BluePyOpt provides methods for setting up both small- and large-scale optimisations on a variety of platforms, ranging from laptops to Linux clusters and cloud-based compute infrastructures. The versatility of the BluePyOpt framework is demonstrated by working through three representative neuroscience specific use cases.

Parameter tuning differentiates granule cell subtypes enriching the repertoire of retransmission properties at the cerebellum input stage

“…The ionic channels were distributed among dendrites, soma, hillock, axonal initial segment (AIS), ascending axon (AA) and parallel fibers (PF) (Masoli et al, 2017). The maximum ionic conductances (Gmax) were optimized using routines based on genetic algorithms (BluePyOpt) (Deb et al, 2002;Zitzler and Künzli, 2004;Van Geit et al, 2016;Masoli et al, 2017). The optimization was run iteratively to improve models fitness to an experimental "template" through the automatic evaluation of "feature" values parameterizing the spike properties of the template.…”

“…1). The hypothesis that the known set of ionic channels was indeed sufficient to explain the GrC firing subtypes was explored using automatic optimization of maximum ionic conductances (Gmax) (Deb et al, 2002;Zitzler and Künzli, 2004;Van Geit et al, 2016;Masoli et al, 2017) yielding a family of solutions that fit the experimental "template" (Fig. 5C).…”

“…In a recent work (Masoli et al, 2017), automatic optimization procedures (Druckmann et al, 2007;Druckmann et al, 2011;Van Geit et al, 2016;Migliore et al, 2018) were applied to GrC models yielding a family of solutions with maximum ionic conductance values falling within the range of physiological variability. During short current injections (500-800 ms, as used in previous experiments), all GrC models conformed to the canonical firing pattern but, unexpectedly, prolonged current injections (2 sec) revealed a rich repertoire of adaptation properties.…”

Integrating the Allen Brain Institute Cell Types Database into Automated Neuroscience Workflow

“…Neuroscience researchers doing simulation and modeling often use Python as a programming medium (Antolík and Davison, 2013; Davison et al, 2009; Hines et al, 2009; Muller et al, 2015; Van Geit et al, 2016), whereas researchers on the experimental side often use MATLAB (Baek et al, 2016; Delorme and Makeig, 2004; Englitz et al, 2013; Felice et al, 2016; Lawhern et al, 2013; Schrouff et al, 2013; Shamlo et al, 2015; Vidaurre et al, 2011); there are exceptions to both as well as hybrids. On ModelDB as of February 28, 2016, there were 247 MATLAB models and 104 Python models hosted or linked to; NEURON models numbered 523 (McDougal et al, 2017).…”

An efficient analytical reduction of detailed nonlinear neuron models

“…Modern personal computers can simulate tens of seconds of electrical activity of single neurons comprising thousands of nonlinear compartments and synapses. However, they handle poorly cases where many model configurations need to be evaluated such as in large-scale parameter fitting for single-neuron models 5,28 , or when the dendritic tree is morphologically and electrically highly intricate and consists of tens of thousands of dendritic synapses, as with cortical pyramidal neurons 15 . When the aim is to simulate a neuronal network consisting of hundreds of thousands of such neurons, only very powerful computers can cope.…”