Exporting NeuroML from PSICS

Most of the discussion so far has focused on the issues involved in importing models expressed in NeuroML for use in PSICS. The conclusion has generally been that it can't be done automatically but that a fair fraction of most NeuroML models can be imported automatically and that it is not particularly onerous to patch in the rest by hand.

The answer to the corresponding question as to whether models built for PSICS can be exported as NeuroML is generally "no". A large part of the reason for this is that PSICS simply addresses a different problem domain form that for which NeuroML has been developed. If it did not, then there would not be much point in writing it. However, the concepts that can be expressed in PSICS are there because there is a modeling demand for them, so it it is of some interest to enumerate the main features present in PSICS that are currently missing from NeuroML since these may prove to be useful candidates for inclusion in future versions.

The formats reference details the exact forms used for PSICS input files. Here we focus on the general concepts that can be expressed in PSICS models that may be of interest in NeuroML.

Morphology

The essential feature of PSICS morphologies is the ability to attach unique labels to user-selected points in the structure. These are used for defining pionts for recording and stimulation as well as for bounding regions used for allocationg channels.

Two extensions to the core point-based model are the use of Branch elements with an offset attribute to specify branches that emerge perpendicularly from the parent process (typically spines), and the use of points with a beyond attribute instead of an absolute position to indicate that a point extends its parent process by a certain distance in the same direction without the user having to define the exact coordinates.

Channel Kinetics

The standard transition type for PSICS channels is the VHalfTransition which is parameterized in terms of the gating charge z, vHalf potential, vHalf, gating assymetry gamma time constant tau, and either a single saturation rate tauMin or separate forward and reverse reates tauMinFwd and tauMinRev. The transition specification can also include the baseTemperature and q10. There is a functionally equivalent transition specification in the VRateTransition which has forward and reverse rates at zero membrane potential instead of vHalf and tau, but either one is sufficient. These are the only parameterized transitions required for new PSICS models. Any system that supports multi-complex kinetic schemes with one of these transition types should be sufficient for non-deprecated PSICS channels models.

Channel distribution

This is the main focus of PSICS. The majority of the features it provides for specifying channel densities are currently not available in NeuroML.

Stimulation and recording

As far as we are aware this is not covered by the current NeuroML specification. PSICS has a fairly wide range of structures for expressiong recording configurations and stimulation protocols including periodic and repeated steps as well as noise sources and algorithmically defined recording sites.

Units

Almost all specification of dimensionl quantities in PSICS requires the units to be included (the exception is for morphologies where everything is in microns). The standard units set covers most common cases but can easily be extended if required. On export PSICS could easily convert to any desired collection of units but the inclusion of units with a model specification considerably improves its readability and intelligibility.