Complex connections

[Helveg]

Most data formats assume straightforward pre and post synaptic connections, how would MDF deal with more complex cases where synapses can be made up of 3 elements (say a teaching signal for plasticity, cfr classical cerebellar granule-purkinje-IO LTD synaptic mechanism), god forbid 4 to 5 when coupled to glial cells (cfr classical astrocyte tripartite synapse) or neurovasculature?

[jdcpni]

It should be possible to handle this in a straightforward way with Ports. However, we will discuss this — and your other message — in our next team meeting and get back to you with a more detailed reply. In the meantime, thanks for your interest! jdc On Feb 21, 2022, at 7:09 AM, Robin De Schepper ***@***.******@***.***>> wrote: Most data formats assume straightforward pre and post synaptic connections, how would MDF deal with more complex cases where synapses can be made up of 3 elements (say a teaching signal for plasticity, cfr classical cerebellar GrC-PC-IO synapses), god forbid 4 to 5 when coupled to glial cells or neurovasculature? — Reply to this email directly, view it on GitHub<#191>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACSHKGCM6NUBEO3M2BY6LDDU4ITQXANCNFSM5O6M3OEQ>. Triage notifications on the go with GitHub Mobile for iOS<https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android<https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Helveg]

Thank you, I look forward to your reply, and adopting MDF ;)

[pgleeson]

Thanks for your question @Helveg. These more physiological scenarios would certainly be supported by MDF down the line, but it's a bit of a way off before there would be support from them in "pure" MDF. A more likely scenario is expressing such networks in NeuroML, and relying on the automated conversion of NML to MDF, ideally (re)structuring the MDF to ensure they run as efficiently as possible on platforms like PyTorch.

There is some NML-MDF interaction underway here: https://github.com/ModECI/MDF/tree/main/examples/NeuroML, but it's mainly focussed on mapping the internal dynamics (in LEMS) to MDF so far.

See also https://github.com/NeuroML/NeuroMLlite, which represents the latest work on more efficient/compact network representations for NeuroML. This shares some of the underlying framework with MDF and there will be deeper compatability/overlap in future.

[Helveg]

Ok, I hadn't realised that to be the focus of MDF :) That makes my questions a bit moot! They're mostly about scaling biophysical models. I was already wondering about the relationship between NML and MDF