OCNS Infrastructure/software/tools SIG: meet and greet, initial discussions

Hi everyone! I’d like to start an infrastructure/software/tools special interest group (SIG). We can find ourselves a name later, but this is the general idea.

Computational neuroscience cannot exist without the tools that we all rely on. So, this SIG will focus on these. We’ll find and discuss tools, we will learn how to use them, we will test and review them, we will file bugs to inform the developers of issues, and finally we will try to learn how they work and try to get involved in their development—to ensure that these tools that we rely on remain in good shape by having communities looking after them.

While we do all of this, we will improve our knowledge of transferable skills: anything involved in modern software development (which are not limited to only writing tools, but are extremely important/useful when developing models using these tools).

So if this sounds interesting to you, please give a short introduction of yourself—what you do, what your interests are; and follow that up with what you’d like to see the SIG do.

11 Likes

I’ll start:

I’m Ankur. I trained in computing and moved to comp-neuro for my PhD. My PhD project at the UH Biocomputation Group was related to plasticity—structural and synaptic (CNS poster here). I used the NEST simulator, and a lot of Python for my data analysis. I’ve recently started as a post-doc at the Silver Lab at UCL where I work on the Open Source Brain platform, and learn about the Cerebellum and other data-driven projects that the lab undertakes.

Since I come from computing and have a long relationship with Free/Open Source software as a Fedora project contributor, I like to see well written code that is accessible to all, and I like to see communities develop around software projects to help maintain them together. An example is the NeuroFedora project that is completely run by volunteers—most of who aren’t even in neuroscience.

So, I’d like to see the software SIG help maintain/improve/develop/disseminate the tools comp-neuro depends on.

2 Likes

I too would be interested in a software SIG. I am a computational scientist and the creator of the Brain Dynamics Toolbox. I believe that scientific tools should be both intuitive and flexible. However it takes a lot of design iterations to achieve that and the effort often goes unnoticed. A SIG could help spread the word and maybe even help attract funding to support the work.

  • Stewart Heitmann
3 Likes

Hi! I am Caglar, and I too would appreciate a SIG for tools and software. I am a CNS PhD student at TU Berlin and the Bernstein Center and a developer of neurolib which is a Python framework for simulating whole-brain models using a growing set of neural mass models, including a mean-field model of AdEx neurons about which I have also presented a poster at CNS2020.

neurolib allows you to easily setup a brain network model from DTI tractography, can simulate BOLD activity to compare simulations to fMRI, to do fast parameter explorations and optimization using evolutionary algorithms. We are actively working on the development and are using neurolib in our own daily research. It would be great if more users would explore the library and report issues and tell us about their use cases.

I am also working on a tiny parameter exploration toolkit named mopet, which I think, addresses an issue that many people will have encountered during their research, namely running a large number of simulations on a local machine or a cluster network. mopet relies on ray, which manages the distribution of jobs across different nodes.

I’d be happy to be part of this SIG and I’m looking forward to more interaction with other projects in this space.

4 Likes

Hi all! My name is Nikola, I am as well interested in tools and software for CNS. I am currently a postdoc at TU Berlin with background in physics and causality detection. I am helping Caglar with neurolib, in particular, I am interested in heterogeneous network modelling. That is that each node in a network model can be different. By different I mean it can be different model (Wilson-Cowan, Wong-Wang, Hopf model, mean-field model of AdEx neurons, etc), you can couple e.g. thalamus or hippocampus mass model with cortical nodes, each node can have different number and type of masses (e.g. some with one inhibitory mass, other nodes with two inhibitory masses), different parameters, intrinsic connectivity etc.

My goal with this is to allow for very fast prototyping of novel heterogeneous whole-brain models, where you simply define your masses (dynamical equations), nodes (which masses constitute a node) and full network (nodes and their connectivity and delays) and everything else is taken care of by our framework.

This is closely related to my current research topic which is integrating thalamic submodule into a whole-brain cortical network and to study the interaction of cortical slow oscillations and thalamic spindles in this sleeping brain model. I also presented my first results on a poster at CNS2020.

I also developed python code for optimising large-scale biological models via Bayesian technique called GPSO (pyGPSO), where you use Gaussian Process as a surrogate for your objective functions, hence you save costly runs of the model.

I’d love to be part of this SIG to stay up to date with current state-of-the-art options for modelling and to provide my bits into it <3

Take care!

4 Likes

That’s great. We already have some interest :clap:

We’ll wait some more—until after we’ve sent out an e-mail to the OCNS membership about the SIG to begin discussing initial plans. Please spread the word in the meantime, and encourage colleagues to join us.

Hi everyone,
my name is Christoph Metzner and I am also interested in tools and software for CNS. I am a postdoc at TU Berlin and work on models of sensory deficits in psychiatric disorders, from biophysically detailed microcircuit models up to abstract whole brain models. I have contributed (very little) to neurolib as well as the Open Source Brain. Furthermore, I have developed a Python package for automated validation of and comparison of models of auditory steady-state response (ASSR) deficits in psychiatric disorders (ASSRUnit; paper here).

Best,
Christoph

2 Likes

Hi everyone,
My name is Felix Kern, I’m a postdoc at the UTokyo International Research Center for Neurointelligence. Among other things, I am interested in closed-loop experimental approaches with computational components. Over the past several years, I have maintained and extended a dynamic clamp software (StdpC), and built a closed-loop model fitting method (poster) which, while technically open source, is a barely usable mess. I have links to the GeNN development team and have recently started contributing to the brian2genn project.

I would be happy to contribute to discussions, and to testing and code-contributing efforts in the spirit of collaboration and learning.

1 Like

Thanks for your interest @kernfel @ChristophMetzner @caglarcakan @nikolajajcay @heitmann.

The OCNS Board discussed the general idea of SIGs at the meeting on Friday and were very happy to see the interest here already. The next step is to e-mail all OCNS members to ensure everyone is aware of the SIGs, and then to announce us on the mailing lists etc.

Then we’ll start organising ourselves to see where and what we want to start with. Does that sound OK? Please continue to reply here with any comments/ideas/feedback that you may have.

Hi everyone,

My name is Shailesh Appukuttan. I did my undergrad in Computer Science, and then moved on to Biomedical Engineering for my Masters, and then finally into Neuroscience for my PhD. Currently, I am a postdoctoral researcher in the Neuroinformatics Group (led by Dr. Andrew Davison) at UNIC, CNRS (Paris) working on the Human Brain Project. I am interested in the application of computational techniques to biological research. My work involves the design and development of a model validation framework for neuroscience, and its integration into existing model development workflows. Alongside, I am also involved in the development of PyNN, a simulator-independent language for building neuronal network models.

I think its a really interesting idea to initate a SIG for software tools & infrastructure, and holds a lot of potential. One of the first things that I propose we could do is to have a crowd-gathered listing of the various neuroscience tools that are already out there. This would be a constantly evolving list, and every entry would require some minimal metadata (such as purpose of tool, webpage for downloading, OS comapatibility, link to docs, license info, …). This listing of tools should be categorized appropriately into meaningful sections, e.g. simulators, parameter optimizers/fitting tools, analysis tools, clients for handling services, and so on. Each tool can have its dedicated page, which could give a short overview, example, and also provide a discussion forum for Q&A regarding the tools and their usage (I realize that some of this info might already exist for certain tools, e.g. in their GitHub issue trackers, but I am imagining more on the lines of the NEURON forum). Over time this could evolve into a really useful source of info regarding various tools. The SIG team could (depending on available volunteer-force) further “curate” the various tools in various ways.

I guess there is loads we can do, and the above was just the first thing to pop into my head. This is a great initiative, and we should all make it happen.

1 Like

Hi Sanjay,
I’m Zohreh and your idea sounds interesting to me, but I don’t know if I can fit with your expertise or not. I have master degree in medical physiology and my research was about tremor and endocannabinoids. After graduation, I worked as R&D researcher in an electrophysiological company and I’m skillful in variety of extracellular recording and electrophysiological techniques(EEG, ERP, ECoG, EMG, ECG, field potentials, single-unit recording and neurofeedback). Moreover, I was product manager in a startup. My knowledge in computational neuroscience and programming is basic compared to you, so don’t know if I can join your team. please let me know about this. Thank you

1 Like

Agreed. I would like to see a list of computational neuroscience tools too. The SIG could play an important role in dissemination. There may be some overlap with INCF in that regard. Does anybody know?

2 Likes

Of course you can. SIGs are just a bunch of people with common interests who “hang out” together to do different things that match common interests/goals. No prior knowledge of anything is required. Everyone is welcome. A primary goal of SIGs is to enable “knowledge exchange” :slight_smile:

2 Likes

There already are lists of computational neuroscience tools. This list is a great one:

The primary issue with taking on the task of maintaining one is that it needs to remain up to date to be useful, and that of course requires an active team of “curators”. Aiding existing lists may be more useful (and less work)?

I have a slightly different idea where the SIG comes up with a checklist of best-practices for software and then members of the SIG use this checklist to review tools we use. We will then get in touch with the developers with suggestions on improving bits that didn’t pass our checklist and where possible, we’ll help them do it. I’m not aware of a group doing this at the moment.

Most Linux distributions have software guidelines, so we can derive ours from them. Fedora (RedHat based) has quite exhaustive guidelines here (since software in Fedora ends up in later releases of RedHat Enterprise Linux) for example:

https://docs.fedoraproject.org/en-US/packaging-guidelines/
https://docs.fedoraproject.org/en-US/packaging-guidelines/ReviewGuidelines/

A review is just a peer-review. Here’s one I’d submitted for Neuron as an example (this can all be done over GitHub):
https://bugzilla.redhat.com/show_bug.cgi?id=1662526


We should timebox things now. I’ll poke the OCNS Board about e-mailing all our members, and I’ll e-mail the mailing lists today.

The primary task at the moment is to set up our communication channels. So:

  • for asynchronous discussion: is Neurostars OK, or do we want a mailing list (infra-sig@cnsorg.org can be set up easily for example)
  • for synchronous discussion: what do people think of a Gitter channel? It’s open, public, and can manage large numbers of people chatting. (I’d like to avoid Slack: it’s proprietary, and is invite only).

This is something we need to look into. The hope is that someone that works with INCF will also work with us, and thus serve as a bridge between the two.

1 Like

The Center for Reproducible Biomedical Modeling have similar aims. PLOS Computational Biology are piloting them as part of manuscript reviews. https://reproduciblebiomodels.org/
https://biologue.plos.org/2020/05/05/improving-reproducibility-of-computational-models/

1 Like

Happy to hear that. Let me know the probable procedure.
Moreover, I forgot to mention my research interest: I’m passionate about time function in the brain and also neurofeedback (enhancement of this techniques by working on its feedback modalities and also finding neuromarkers for disorders like depression and anxiety, ADHD subtypes and …).
Thank you for gathering us together.

1 Like

I like your idea about reviewing tools based on best practices, and having developers looped in on how things can be improved. Alongside, a good part of our focus should also be directed towards publicizing existing tools, and educating the community about these. A listing such as the GitHub page (https://github.com/asoplata/open-computational-neuroscience-resources) is useful in what it does, but very limited in what if offers. For sure, we would require a lot of volunteers/curators to keep any of this upto date. So we would need to see how many signup to be a part of the SIG. I would be very keen to see the SIG help disseminate tools by conducting periodic events, such as bi-weekly online sessions where the authors of a particular tool have a short (~30 mins) presentation and demo about their tool. The recordings of these can then be made available on a OCNS SIG YouTube channel for even future users to get a quick glimpse of the various tools.

1 Like

Periodic meetings is certainly something we should try to do.

Just FYI: I mailed the comp-neuro, systems neuroscience, and connectionists mailing lists about the SIG today.

2 Likes

Hi ! I’d like to join the group too.
My name is Amelie Aussel and I’m currently a postdoc at Boston University. I’ve been using Brian2 for Python for about 4 years (since I started my PhD in France), mainly to build Hodgkin-Huxley based neural networks. My current project aims at modeling attentional processes. I also have an engineering background.
I’d be very happy to learn more about the many tools available and how they were developed, and discuss potential new features :slight_smile:

1 Like

Hi, I am Chaitree Baradkar. This looks like a great initiative. I am not sure how can I contribute though as I am not from computational neuroscience background. I have been working in Machine Learning for 4 years and very interested in interplay between machine learning and computational neuroscience. Please let me know if I can be of any help.

1 Like