Electrical stimulation of the nervous system can treat a range of diseases including epilepsy, heart failure and chronic pain. However, these therapies are limited by inadequate stimulation of targeted (therapeutic) nerve fibers and unwanted co-activation of non-targeted (side-effect) nerve fibers. We developed, validated, and shared computational models, code, and data that enable simulation of electrical stimulation of nerves, and these resources have found widespread use and impact, as indicated by downloads and citations. The models enable the continued advance of bioelectronic therapies by assessing the effects of stimulation parameters and enabling patient-specific optimization to accommodate individual differences in neural response.
Our work developing and deploying computational models of neural stimulation is conducted by a diverse group – including gender, race, ethnicity, institution and professional level: undergraduate, Masters, and PhD students, postdocs, staff scientists, and Asst to Full Professor. We collaborate with Ezzell and Clissold at UNC to quantify nerve morphology, providing essential inputs to the models, and with Ludwig, Settell, and Blanz at UW to validate our models with in vivo data. Overall project leadership and mentorship are provided by Grill & Pelot, and Pelot leads our efforts in model and data sharing. We updated our data, metadata, and protocols per NIH SPARC standards, including internal peer-review, followed by SPARC curation before publication. Pelot serves on the Change Control Board for the NIH SPARC Consortium to provide user-side feedback on data curation and sharing processes. We use git for version control and coordination across multiple developers. Team collaboration and data management tools include folder structures on Box and a wiki for sharing protocols and resources. Thus, in addition to developing tools to advance bioelectronic therapies, we are training scientists and engineers in modeling and data sharing.
Clinical outcomes of neural stimulation are mixed and pre-clinical results are not predictive of successful translation. The simulations enabled by our models, code and data allow scaling of parameters from animal to human, design of patient-specific therapies, and optimization of stimulation parameters.
We shared our first model on NEURON ModelDB in 2002, and although this site does not track downloads, the paper has >775 citations. The model has been used in diverse applications, from studying the degenerative effects of multiple sclerosis, to designing interfaces to restore sensation to amputees, to analyzing brain stimulation to treat Parkinson’s disease.
In 2021, we shared our ASCENT pipeline to enable non-experts to implement & analyze highly realistic models of nerve stimulation. Within only 1 yr, our GitHub repository received >500 unique visitors and >100 unique cloners. We developed extensive documentation on Read the Docs, and we committed substantial effort to promote accessibility, including the first Success Story on NIH SPARC community feed, a demonstration booth at Experimental Biology 2022, a video tutorial on installation and first simulation, and a ready-to-use installation on the NIH o2S2PARC simulation core.
Implementing nerve models requires defining the morphology of the nerve of interest. For example, the vagus nerve is stimulated to treat a range of diseases, but stimulation of off-target nerve fibers causes side effects. We quantified the morphology of the vagus nerve in rats, pigs, and humans and shared all data, including raw histological images, analyzed metrics, analysis code, summary statistics, animal and subject demographics, protocols (on protocols.io), and summarized historic data from literature. Our datasets follow the SPARC sharing standards, with file organization and metadata based on the Brain Imaging Data Structure (BIDS). These FAIR datasets allow others to [re]analyze our data for replicability and to address their own research questions. For example, these data allow implementation of a virtual population of nerve models to quantify differences in neural responses across individuals. In <2 yrs, the paper has 21 citations and our dataset has 42 downloads.
Collectively, these tools enable researchers to advance new neural stimulation therapies, including translation from preclinical to human nerves, design of patient-specific therapies, and optimization of stimulation parameters.
As we teach our trainees, “just because the code runs and numbers come out, does not mean they are correct”. We follow best practices for rigor and reproducibility in computational modeling(1), including model version control using git and formal evaluation of model conformance using the Interagency Modeling and Analysis Group rubric(2). Our reference materials for team members include “best practices”(3) for scientific computing and key characteristics of robust code(4).
We engage alpha- and beta-testers from our team and at other institutions who have varied levels of experience in programming and neural modeling. We also compare our model predictions against other software; for example, to validate the results from ASCENT, we found strong agreement with neural activation thresholds for matched models simulated on Sim4Life by the IT’IS Foundation (5, Supp 35).
As we continue development of our ASCENT platform, we release new stable versions to the public GitHub repository, with corresponding updates on Read the Docs, and we revisit our beta task to ensure replicability (5, Supp 45). Each release is assigned a DOI via Zenodo to ensure that publications using ASCENT can specify version.
1 Stodden et al. 2016 Science 354:1240
2 Erdemir et al. 2020 J Transl Med 18:369
3 Wilson et al. 2014 PLoS Bio 12:e1001745
4 Benureau & Rougier 2018 Front Neuroinform 11:69
5 Musselman et al. 2021 PLoS Comp Bio 17:e1009285
Data sharing and reuse accelerate the pace of scientific progress by increasing rigor and reproducibility, reducing redundancy, creating opportunities to glean new insights from data, and, for us, have provided substantial impetus for new and productive collaborations. We shared a published dataset with mathematicians led by K Flores at NCSU, and this collaboration resulted in a new method to detect urinary incontinence and a peer-reviewed publication. Our efforts to model nerve stimulation experiments conducted by K Ludwig at UW involved exchange of data and models, and this led both groups to document more extensively the experiments and models. We used our simulation tools in collaborations with multiple medical device manufacturers, including LivaNova, Boston Scientific, and The Alfred Mann Foundation, to analyze and develop next generation neural stimulation technologies.
Dr. Pelot was one of two panelists for a NIH HEAL Initiative “Fresh FAIR Webinar”. Such conversations are key to shifting the scientific culture and expectations around data sharing; this collective effort will benefit current and future researchers and will lead to faster and more effective treatment of diseases.
