Advanced metamodelling on the o2S2PARC computational neurosciences platform facilitates stimulation selectivity and power efficiency optimization and intelligent control

o2S2PARC (Open Online Simulations for SPARC) is a cloud-based, web-accessible andcollaborative platform for open, FAIR, and reproducible computational neurosciences,developed in the context of the NIH SPARC program (Stimulating Peripheral Activityto Relieve Conditions). It supports mechanistic (e.g., biophysical) and data driven(e.g., machine learning) modelling and the generation of complex pipelines andworkflows for a large number of preexisting or user-generated modules, e.g., for personalizedtreatment planning or device safety and efficacy assessment. We introduce ametamodelling framework within o2S2PARC towards facilitating tasks such as optimization,sensitivity analysis, uncertainty propagation, surrogate modelling, and failure modedetection. A port scheme for the parameterization of pipelines and extraction of quantitiesof interest was introduced and combined with infrastructure for distributed computing,orchestration, and rapid inter-service communication, and with the Dakota library(Sandia National Labs) for advanced metamodelling algorithms. We will demonstrate theapplication of the resulting framework for i) the multi-goal optimization of stimulationpulse shapes for superior effectivity and power-efficacy, ii) uncertainty quantificationfor a neural interface safety assessment pipeline, iii) electrical-impedance-tomography-basedstimulation-selectivity optimization in multi-fascicular nerves, and iv) model-basedclosed-loop control of a bio-electronic implant. A powerful API ensures that themeta-modelling framework can be flexibly and broadly applied. Extending o2S2PARC withsuch a framework provides the neuroscience community with valuable functionality forin silico studies.