180 likes | 191 Views
Discover the current status and future plans of The Web Lab, a platform that enables reproducible model development, parameter fitting, and comparison between experimental datasets. Explore the potential of automating model evaluation and deploying models in the Virtual Physiological Human.
E N D
The Web Lab: current status & future plansGary Mirams CellML Workshop 2018
An imagined conversation in 2010 Experimentalist: I’ve recorded some action potentials from a dog myocyte at 2Hz. Are your models any good at predicting them? Modeller: Er, let’s have a look at these papers on dog models… Oh. None of them have a 2Hz action potential in them. Experimentalist: Well, where can we look them up? Modeller: Er, you can’t. I’ll have to download them all, and write some code / or use a GUI to get you the 2Hz traces. Give me a day or two. Experimentalist: Are you telling me your so-called “most mature field in systems biology” has no way to show me what a 2Hz action potential looks like for your dog models? Modeller: Er… afraid so.
Motivation C = closed O = open I = inactivated Different published structures for Ikr models (cardiac potassium ion current)
Different models, different predictions Some variation expected, but which model should e.g. the FDA use?
Motivation for Web Lab 1 • Comparing model behaviours • Different models in the same situation • Or one model in different situations • Assess suitability for a new study • Record process of ‘model->figure’ for reproducibility
Here are the dog 2Hz action potentials https://goo.gl/NfUVxw
Key features summary • Consistent application of a protocol to any model • Interface described at the level of biophysical concepts • Units conversions are all handled automatically • Specify model inputs and outputs • Simulator works out which equations it needs for that simulation • Replace components • For example encode your own stimulus protocol, or apply voltage clamps, even to alter or add new equations (e.g. change/add ionic buffering to match an experiment) • Includes all the post-processing and plotting instructions • Ability to do complex parameter sweeps, analysis, etc.
Things left to perfect • Annotations – at present our own ‘oxmeta’ simple ontology stored in the CellML files as RDF tags. But we’ve had to copy models from the CellML repo: https://github.com/Chaste/cellmlCan we work out how to annotate the official CellML repo files instead and get started on that? • Protocol language – quite tricky to learn/write, and very tricky to debug.Is it desirable to have such a markup language for protocol definition that includes postprocessing?
What next? A vision of the future Knowledge about mechanisms is captured in quantitative models Best experiments to do are therefore the ones that best [select and] parameterise the model Provide these to experimentalists Automate model development and evaluation of predictive power Deploy in the Virtual Physiological Human!
Motivation for Web Lab 2 Reproducible Model Development: https://doi.org/10.1016/j.pbiomolbio.2018.05.011 Experimental data Parameter fitting Technology stack revamp
Web Lab 2 – adding data! • Will also be able to compare between experimental datasets • Ontology-based search and selection
Pints PINTs – Probabilisitic Inference for Noisy Timeseries https://github.com/pints-team/pints A back end to do the fitting aspects, and log what has been done into a Web Lab fitting spec.
Acknowledgments • Project team • David Gavaghan • Jonathan Cooper • Gary Mirams • Michael Clerx • [Aidan Daly] • Asif Tamuri • Helen Sherwood-Taylor • Collaborators • Steve Niederer (KCL) • Rick Gray (FDA) • Kylie Beattie (GSK)