For rule 1: The goal of the project is to predict spatio-temporal distributions of active neurons based on current densities created by multi-electrode electrical stimulation, and dependent upon a set of "core models" of molecular (receptor-channel kinetics), synaptic, neuron, and multi-neuron activity. The model is intended to be used for (1) theoretical research that links multiple physical scales to network activity or a cognitive process (e.g., spatial navigation or memory), (2) electrode and stimulus design, and (3) drug discovery.

For rule 2: All model parameters were constrained using experimental data where available, e.g. unitary postsynaptic potentials for the appropriate pre-post neuron pairs, neuron electrophysiology specific for a neuron type, axon distributions, etc. The models validated against the appropriate experimental data, e.g. evoked potentials due to extracellular stimulus. These data and their references are described in all relevant publications. In certain cases, model development using prior published models, not from our research groups, as starting points and are referenced in publications. Their structure and parameters values are available in their respective publications.

For rule 3: All models are evaluated within their respective contexts and corresponding levels of granularity (See List of Planned Activities Table, section Systemic Calibration and Evaluation). For example, parameters for our abstract (data-driven) models derived from mechanistic models are estimated while minimizing error between output from the mechanistic (detailed and computationally complex) model and output from the abstract model using a 'training' dataset; validation is performed using a separate 'validation' dataset. 

For rule 4: Mechanistic models drive their predictive powers from the test protocols with which they were initially calibrated and tested; significant variations in the simulation protocols may consequently affect model predictability levels. The specific protocols using which the models were constrained are described in publications; further limitations are described in the discussion sections.  In addition, for models already shared with the community, these publications are explicitly mentioned on the webpages on which the models are shared. 

For rule 7: Detailed models out of prototyping phase and used for development of input-output models were published and their code shared on open-source database (ModelDB).  IO model structure was also described in multiple publications.  Other IO models will be similarly shared as they become available. Notably, the admittance/neuron bridge model will be made accessible within the next year.  The ROOTS algorithm is already freely available on GitHub and through the PyPi library. 

For rule 9: We compared our IO models to multiple IO and mechanistic implementations, which helped us quantitatively determine our model's capabilities and limitations (details in publications). The starting point for our mechanistic neuron models are previously published models which are cited in our publications. In adapting these models to our platform, tests are performed to verify that our models are capable of recreating the figures associated with the original models. At the network level, there are yet to be comparable models to which we can compare. This is the same for modeling the effects of extracellular stimulation.  Notably, our efforts to discretize the extracellular space led to two meshing methods (hexahedral and tetrahedral) which enabled us to mutually validate their results.  

For rule 10: Our models are developed using best coding practices and operating procedures. Notably, the neuron and network models do not yet conform to a standard mode description language such as NeuroML. We plan on converting the code to this standard by 2021. On the other hand, the mechanistic models of biomolecular processes are all coded in the SBML standard (Systems Biology Markup Language).