Skip to main content

Making a model: Part 6 - manual changes (RyR)


Did everything we attempt to improve end in success? Alas, no. One thing I don’t particularly like about the ORd’s calcium handling is the fact that the release from the sarcoplasmic reticulum (SR) depends directly on the L-type calcium current. The release should ideally depend, among other things, on the calcium concentration adjacent to the sarcoplasmic reticulum. Beyond the phenomenological nature of the dependence of release on L-type calcium current, this for example also prevents the possibility of the cells manifesting delayed afterdepolarisations in a physiological way [1].

Basically after finishing the ToR-ORd model as it is now, Dr. Alfonso Bueno-Orovio has suggested that calcium-dependent SR release would be a really nice thing to have, with which I heartily agreed, and we started working on it together. I really needed a break at the time, but then again, we thought that if it was achievable to change the SR release without breaking many other properties, it would be silly not to try it. There aren’t so many models of calcium-dependent SR release, so we experimented with several formulations. Unfortunately, with regards to what phenotypes the model could show, none worked as well as the existing formulation. It was often impossible to get reasonable alternans [2], and in some cases the rate dependence of calcium cycling seemed to become rather exaggerated (huge increases in SR loading with increased pacing rate etc. [3]). Anyway, we tried many different things together with Alfonso – first it didn’t look promising, but we were in such a situation many times before, and there were many things on the TODO list to check and explore if they couldn’t remedy the issues. Except one day, the list became empty – we explored what we could, perturbed all the parameters, tried various formulations. We looked at each other one day and almost in unison agreed we’d give up.  I suspect the issues we met are probably linked not just to the SR release, but to other elements of calcium handling in the existing model: SERCA pumps, CaMKII, SR diffusion, calcium buffering, maybe the cell compartmentalization, and potentially also the L-type calcium current. I think these parts of the model are in some form implicitly adjusted to work nicely with the L-type-calcium-current-dependent SR release, and when we want to change the SR release, the other components might have to change as well. Doing such changes would be another major project in itself, and we felt it was out of scope of our existing work. The issue with model development is that you know how long it takes only when you finish – there is no guarantee that hard and meticulous work alone will give you success. For such reasons, we drew a line and agreed we’d keep ToR-ORd as it is.

It is an interesting philosophical question on how much to prefer the set of phenotypes that a model manifests consistent with literature, versus accurate mechanisms. I generally lean towards the mechanisms - whenever there was a situation in ToR-ORd development, where I thought “well, this is a bit off mechanism, but seems to work fine for now”, it always came back to kick me and it had to be reworked more thoroughly. On the other hand, there has to be some balance, and it is hard to persuade others that a new model is really worthwhile when it seems, at the first sight, to behave worse than previous ones. People are busy and the first sight may be the last one they give. My first intuitive reaction to such thinking is “but you don’t want to have a good behavior with the wrong mechanism”, but then I’m afraid that models are models, and for most mechanism there (if not all), one can go deep enough to find some flaw. Rather than to go for the wild goose chase of trying to fix everything, it may be perhaps more reasonable to be well-aware of the model’s limitations and take them into account when interpreting model simulations. This is by the way another reason why I’m so much in favor of deep model analysis that goes to the low-level of the models workings, as only then it’s possible to assess how plausible the model prediction is. A newly discovered ionic current/flux is incorporated in a model and it induces a particular behavior? Great, but why exactly? Computer simulations (as is often stated in article introductions) allow perfect observability and control over the model, so why not to use that?


[1] It is possible to come up with some form of explicit store-overload mechanism depending on the SR loading. However, this then may not be very sensitive to established important factors in formation of delayed afterdepolarisations that rely on elevation of calcium adjacent to the SR, such as SR leak etc. 

[2] Couldn’t we be happy with a model without alternans? We probably could – however, unlike some other prearrhythmic behaviours which are relatively sparse and usually require disease or drug block to manifest, alternans appears to be achievable relatively consistently even in healthy hearts. Even if there are exceptions (which can be also due to laboratory protocols), capability of alternans formation still seems to be relatively general property of myocyte calcium cycling and losing it raises a flag that something is not ideal.

[3] At the same time, many properties were not broken and in most regards, the model was absolutely fine, giving us hope and even a bit of belief that the ToR-ORd model is generally sensible.

Comments

Popular posts from this blog

Several tips on junior fellowship applications

It turns out I was fortunate enough to receive the Sir Henry Fellowship from Wellcome Trust. This is a four-year international fellowship that will allow me to spend some time at UC Davis, California as well as University of Oxford, focusing on interdisciplinary investigation of diabetic arrhythmogenesis. It was certainly an interesting and person-developing experience (obviously viewed more favourably given the outcome). I had the advantage of working under/with highly successful people who gave me valuable advice about the process and requirements. I am quite sure that  I would not have gotten the fellowship without the support of Profs. Manuela Zaccolo, Blanca Rodriguez, and Don Bers, to whom I'm deeply grateful. However, not everyone has such nice and investing-in-you supervisors and beyond very generic advice, there is very little information on the internet on what the process of applying for junior fellowship entails [1]. The aim of this text is to share some findings I ma...

Making a Model: Part 0 - Introduction

Welcome, dear reader. This is the start of a short series of blog posts aimed at providing some insight into the process of development of a computational model of a cell. The type of the model we’ll focus at is one which simulates the development of ionic concentrations and behavior of ionic currents and fluxes over time (probably most relevant for excitable cells such as cardiomyocytes or neurons). I'm hoping that tips and observations in this series will be of use to graduate students and researchers who are interested in computer simulations. While the posts are about the development of human ventricular cardiomyocyte model ToR-ORd ( https://elifesciences.org/articles/48890 ), I mostly try to focus on general observations (usually those I wish I knew about when I started). I decided to write up the topics in the form of blog, given that scientific publications tend to have a somewhat rigid format, and tend to focus at what is done, how, and what it means, rather than at ...

Making a model: Part 1 - Development strategy

This is just a short post about the criteria that one sets for the model to fulfill when making a model. In our paper, we decided to strictly separate criteria for model calibration (based on data that are used to develop the model) and validation (based on data not used in model creation that are used to assess it after the development has finished). Drawing a parallel to the world of machine learning, one could say that calibration criteria correspond to training data, while validation criteria correspond to testing [1] data. In the world of machine learning, the use of testing set to assess model quality is firmly established to the degree that it is hard to imagine that performance on training set would be reported. Returning from machine learning to the literature on cardiac computer modelling, it becomes rather apparent that the waters are much murkier here. Most modelling papers do not seem to specify whether the model was directly made to do the reported behaviours, or wh...