The ISBGroup Blog

So, now
vacation time is over here in Sweden – and work is back with its full
hustle-bustle frenzy. I (Gunnar) started working on Wednesday last week, and
since then I have already submitted one application (to SSF’s industrial Ph.D. programme together with AstraZeneca,
on the creation of a new systems pharmacology platform, based on modelling which is used to
bridge between pre-clinical organs-on-a-chip data and human/clinical
studies), had two supervision days, changed office (my main office is now at
the Department of Biomedical Engineering, and not Clinical and Experimental
Medicine), did the final preparations for the half ironman race this weekend, and went to
Gothenburg to give a lecture in the group of Patrik Rorsman and Charlotta
Olofsson. It is about this last topic, my presentation, that I want to say a
few extra words also here.

Figure 1: Front page of my presentation earlier today.

The title
of my presentation was “Doing biology without modelling is like driving without
a safety belt – it might work, but it might also go really, really wrong”. The
talk was a salespitch to an essentially all experimentalist-audience, and it is
based on an image that came to my mind just a few days ago. The image is a response
to some of the most common forms of critique I usually hear against the usage
of modelling: “I could have said that without using the model”, “I don’t believe
that a model can do everything, I think there is still too much that we don’t
understand”, and “the model only seems to provide an extra degree of confidence
on a conclusion I would have said anyway”.

And, the
thing is that I actually – to a large extent – agree with all of those
statements. Modelling cannot do everything, and should not be oversold – but it can do some things, and those things should be properly appreciated. Similarly, many modelling results are conclusions one could have drawn without
the usage of modelling, and what the modelling does is therefore in many respects primarily
to put and extra degree of confidence on that conclusions, if that was the conclusion you anyway would have made. And just like for a safety
belt, when driving: often you don’t need it, you would be fine anyway, but it
is more secure to have it there, to bring an extra degree of security and
confidence to the current situation. In other words, if a model agrees with
your conclusion, you can be more sure that you are correct. However, in the
presentation, I also gave several examples of cases where the modelling
provides a conclusion that seems perfectly reasonable once you see it, but
where the prevailing conclusion before the model-analysis was done, actually
was a very different one. Apart from our main diabetes examples, I pointed to three probably less known such stories:

Figure 2: The difference between the new way of calculating EC50 values (black) with the old one (red).

1) Our EC50
story, published in FEBS J last year. There we showed that a simple model-analysis could detect a problem with
a previous way of calculating EC50 values: that the steady-state was not
reached in-between the changes in stimuli. In other words, the resulting curve was not an increase in steady-state values, but a long transient overlaying of overshoot responses. That
new interpretation of the data had some predictions, which we verified in independent
experiments, supporting our new interpretation of the data. Furthermore, it was
experimentally not possible to modify the protocol, to wait as long as one should to reach steady-state. Therefore,
the only way for this system to get correct EC50 values (corresponding to steady-state values) is to do what we
propose: to fit the model to a transient responses like the one already existing, and
then use the model to simulate the experiment as it ideally should have been done to start with. As you see in Figure 2 above, the new, more correct EC50 value is almost completely non-overlapping with the old one.

2) An
earlier story of muscle metabolism, where we showed that a seeming
contradiction and missing link – which had been investigated for 25 years – in fact
was not a contradiction at all, but merely a mis-interpretation of data. And
that re-interpreted version sounds very reasonable once you see it. In other
words, for 25 years, people had believed that a not yet discovered regulator of
glycolysis was active in anaerobic muscle recovery, but our modelling showed
that no search for such an unknown regulator is necessary: a correct analysis
of the data shows that the conventional regulators are sufficient to explain
the observations. This story is not yet published, but anyway available as
chapter 11 in my Ph.D. thesis.

3) Another
recent story on interpretation of data for the IL1beta analog Anakinra (Palmér et al, CPT Pharmacometrics Syst Pharm, 2014). In this
story, we had a look at data that seemed to be too good to be true, and
therefore were disbelieved by many: that Anakinra could have lasting positive
effects on the diabetes readout HbA1c as long tie as 1 year after the start of
the treatment, even though the treatment itself only lasted for 3 months. We
showed that a simple model based on pre-clinical data alone did actually
produce that clinical output as well. In other words, we showed that the
initial response was that the initial disbelief in the clinical data was unnecessary: they were perfectly aligned with the pre-clinical data.

Figure 3: Last page in the presentation, summing up the main advantages of doing modelling, i.e. pointing out some of the drawbacks of not using it.

These are three examples that show that
modelling in principle does not do anything that one anyway does: analysis
of experimental data to draw conclusions and suggest new experiments. However, modelling does these things in a more reliable fashion, and it is easy to go wrong otherwise. In other words, to do biological data
analysis without modelling, is just like to driving without a safety belt: it
might work, but it might also go really really bad – throwing away 25 years
of your life.

Figure 4: On the train on the way back, I happened to be seated right next to a very nice systems biology colleague of mine: Adil Mardinoglu. He told me that he and some colleagues of him had read my last blog post, and wanted to contact me about it. So that is actually the reason why I got inspired to write a new one as well!

A couple of weeks ago, our M.Sc. student William Lövfors presented his M.Sc. thesis entitled: “A first phosphoproteomewide mechanistic model of insulin signaling”. As the title indicates, this thesis presents the – to our knowledge – first ever presented version of a dynamic mechanistic model (based on ordinary differential equations and realistic assumptions of protein-protein interactions) that describes the entire phosphoproteome, i.e. all phosphorylations in all proteins that are relevant in a certain scenario. The scenario that we study here is the scenario that has been studied the most in our group, insulin signalling, and this systems-wide model is therefore an extension of our previous models. In fact, our most well-developed previous model appears as a sub-model – as the “core” model – in this new systems-wide model, which has been constructed in an iterative fashion, where we add protein after protein, in layer after layer, where each interaction is taken from suggestions in databases containing known or suggested interactions, and where all phosphorylated states in the model can describe dynamic mass spectrometry data. This work will be continued and experimentally validated and refined in various ways, and hopefully published during 2017. If you want to see more of how far we have gotten up until now in this latest status-report, here is a link to William’s thesis.

A graphical depiction of the model of all added proteins. The old model we had previously developed is depicted in yellow, all proteins that are connected by high-confidence interactions (3 or more references) are depicted in dark blue, and all proteins reached using one or several low-confidence interactions are depicted in light blue. All proteins can describe time-course data.

A final comment can be made about the academic journey of William so far. He started his journey, as so many others, by doing our project course in the year 3 of the engineering biology programme (TB), whereafter he did a ~9 months intership combined between ISBgroup and Mika Gustafsson’s group (then a sub-group of Mikael Benson’s group). His project then was devoted to a related project, where we created a systems-wide ODE model for gene-gene interactions (submitted to PLoS Comp Biol). Thereafter, he stayed in contact and in projects with us during the remainder of his M.Sc. studies. The initial project was financed as a scholarship-based research preparatory course, and he has been funded by some occasional months here and there for work in projects and for supervision in our project course, but mostly just by the fact that William has enjoyed working in these projects. During this time he has published one scientific paper (showing that you sometimes need mathematical modelling for something as basic as a correct calculation of EC50 values, link here), has done an oral presentations in international workshop (ISGSB 2014), and is scheduled to be co-author of two additional papers: one soon-to-be-submitted on adiponectin secretion, and one based on the systems-wide insulin signalling study already started in his thesis.

All in all, such an impressive CV is unusual for somebody who has just a few weeks ago finished his M.Sc. thesis, and we are therefore proud to say that William will stay on in our group as a proper employee during the next 6 months, during which we plan to convert his position to a proper Ph.D. position.

Picture of William, also used on his personal home page, here at ISBgroup.

Spara

A new travel period is just completed. The first part of this travel period can in many ways be described as the travel of 2: 2 continents were visited (North America and Europe), it took 2 weeks and 2 days (June 14-30), two research groups were visited (Chris Sander’s and Markus Covert’s), two companies were visited (Merrimack and AstraZeneca), two scientific conferences were attended (Systems biology of human disease in Boston and BioSynSys in Bordeaux), and two lectures were held (at BioSynSys and Merrimack). This totalled a distance of approximately 22 000km (Figure 1).

Figure 1: Outline of the first part of the travel.

The travelling in the US generally went to Harvard and MIT in Boston, and to the universities in California (Stanford, UCLA, etc). These visits mark a new phase in our groups development, which just has begun: one where we focus and aim more for high-impact publications, and one where we start to focus on making a difference in society.

Figure 2: The methods we have developed, first in the thesis of Cedersund (left), and then by their application in end-usage projects (right). The idea is to combine a mechanistically detailed grey-box model (the box), with a well-determined core, a combination we refer to as a core-box model, which also is our group logo.

In previous and now more-or-less completed phases in our group development, we have had a series of other focuses, which still exist, but now have been transcended. The first focus, in the Ph.D. thesis of the group leader Gunnar Cedersund, was to adopt the data-driven methods in the system identification community (pioneered by e.g. our local legend, Lennart Ljung), to work also for mechanistic modelling of biological systems. This involved the creation of an embryo to our currently used core-box modelling framework, which consists of detailed mechanistic models (the white box in Fig 2) with a well-determined core (the sphere in Fig 2), which pinpoints the uniquely identified predictions. The second phase in our development was to establish a strong and integrated link with one experimental group, to learn to do truly integrated work, where modelling really makes a difference, and answers real, interesting biological questions. This resulted in a series of papers devoted to insulin signalling, where we first learned to use hypothesis testing (Cedersund, 2010) to systematically unravel a small system consisting of only two proteins (Brännmark, 2010), then learned how to map a simple intracellular system with the whole-body level in multi-level hierarchical models (Nyman 2011), and finally learned how to unravel network malfunctions in diseases (Brännmark 2013) (Fig 3). This process is still ongoing: we are still using our most recent version of core-box modelling (Cedersund, 2012) to add pieces to network, such as Erk and Elk (Nyman, 2014), fatty acids (Sips, 2015), and FOXO1 (Rajan 2016), and are also preparing for new quantum leaps to yet more comprehensive modelling levels.

Figure 3: Our multi-level model for insulin signalling and the induction of insulin resistance (red T2D inhibition of the green feedback signal), which also propagates to the whole-body level.

However, since Cedersund recieved a start-up FoAss project grant from the Swedish Research Council in 2011, he has expended this recipe to a third focus: to apply these new tools to a more wide variety of biological problems. This focus has resulted in the joint recruitment of 5 new Ph.D. students, which are jointly employed with other researchers, who do experiments on other systems, but who cannot themselves do the modelling. The general recipe for all these types of collaborations are that the students sit a few days of the week in our group, and the other days of the week in the experimental group (Figure 4).

Figure 4: Our group structure, we are in the middle, and our collaboration partners, providing data and biological expertise have a shared environment with us, in terms of shared students, who sit partially with us, and partially in the experimental environment.

This phase is still ongoing, and we now have more than enough of experimental collaboration partners, and these projects are also starting to produce papers on e.g. modelling of the liver (Forsgren, 2014), the brain (Lundengård, 2016), who use mechanistic models as a means to interpret imaging data in order to produce a new level of understanding and also new biomarkers that can be used in diagnosis of liver and brain diseases (Figure 5). All of these developments mean that we now know how to both methodologically and socially interact with a wide variety of experimental groups to produce a steady flow of papers.

Figure 5: Examples of other application areas, where we combine magnetic resonance imaging data from the brain (A) or the liver (B), with mechanistic models (green below), to identify model properties which serve as new biomarkers. These biomarkers are combining the information in the data with the prior process understanding, and typically outperform existing biomarkers, in terms of providing a correct diagnosis and patient stratification.

The next phase and focus will therefore be to also establish some more high-profile collaborations, that allows us to move upwards in journals, and eventually publish somewhat regularly also in top journals such as Nature and Science. To do this, we need not only top-notch, integrated modelling, but also top-notch experimental data produced by groups who already have a track-record of publishing in high-impact journals. The highest density of such groups exist in Boston (MIT, Harvard) and California (Stanford, UCLA, Caltech, etc). Therefore, this year I have made two trips to these areas to start to get familiar with these geographical areas, and to gradually start to interact with new groups in these areas. Therefore, a 2 or 3-week trip to these two areas is planned to be an annual tradition. And therefore, we have started to write applications together with groups from these areas. One of these applications has now been approved – a two year paid exchange visit by our junior group leader Elin Nyman – but this is the topic of a separate blog post.

Figure 6: People walking around in the relaxed athmosphere that is Almedalen: on this streets you are equally likely to encounter a top politician as a top scientist, a business leader or an environmental activist or a journalist. They are all active, and all guests and participants in this exciting event.

Figure 7: A little bit further down the same street, I encountered one of the many many small little sessions. Most sessions feature 1-6 lecturers, and between approx 15-150 people in the audience. This allows for informal and many parallel sessions, where one can dive into in-depth and meaningful discussions. In total more than 3000 events are held during the 8 day event, and the events are spread out all across the biggest city of Gotland, called Visby.

The rest of this blog post, I instead want to devote to the final part of the just completed travel period: a visit to Almedalen. As is explained already in a previous blog post, Almedalen is Sweden’s, and potentially even the world’s, biggest political arena, featuring >3000 events in an 8 day festival. This is the first time I and we feature at this event, and also the first time we visit it. I am therefore happy to report that the event was every bit as exciting as I had hoped. There really are hundreds of fascinating talks and seminars and panel debates each day, the environment really is conducive for a relaxed, informal, and productive discussions, and the density of interesting and strategically important people is really so high that networking and chance encounters with strategically important people becomes almost automatic. In short, everybody is there: the most important political leaders, the most important media outlets, the most important organizations and companies, and – I learned now – also the most important scientists. I met and could easily engage with several key scientists that I have been wanting to meet for quite some time, and many of my favourite scientific topics (concerning replacement of test animals, eHealth, decision-support, automation and replacement of jobs, environmental solutions, etc) were brought up in numerous talks and debates, and it is really in the intersection of all of these sectors of society (science, politics, companies, etc) and all of these areas (environment, health care, jobs, etc) that I think that our society truly moves forward. I therefore for sure plan to make also this second half of my travel period, the Almedalen-visit, a regular tradition.

I will write a separate blog post about the particular panel debate in which I participated, and about the general topic on which it was held: the three Rs.

—

Almedalen is Sweden’s without a doubt biggest socio-political event. All of Sweden’s top politician’s are there, as are all major media, and most activist groups, and many generally engaged citizen. We are therefore proud to announce that ISBgroup this year will feature at this event. Our participation happens between 9-10AM on July 4, in Kinbergs plats 8, when we will give one of the introductory backgrounds as one member in the expert panel on research without test animals. (more info here) I, Gunnar, who will be the one participating, am looking much forward to this chance to visit this exciting event for the first time, to be able to attend many other interesting lectures, to meet and discuss with interesting, engaged, and influential people, and to be a part of some interesting on-stage discussions following my lecture.

If you are in Almedalen at the 2016 edition, come check us out! 🙂
/Gunnar

Spara

Two weeks
ago, the largest conference on biosensors in history was held in Gothenburg,
with over thousand delegates presenting work within commercialization of
biosensors, printable biosensors, single cell/molecule detection, lab-on-a-chip,
mobile diagnostics, nanonalytic systems etc. There was however room for mathematical
modelling as well. We presented with our poster a way of getting a requirement specification
for the development of glucose sensors.
Since the development and success of the finger prick, enzymatic glucose
sensor, several different glucose sensors has been developed. Despite the vast
amount of different glucose sensors out there, there is still an uncertainty
around which sensors to use when regulating glucose in the intensive care. The
regulation of glucose in the intensive care Is a bit more complex problem than
glucose regulation for diabetics, which is the market that most glucose sensors
aims at. A more complex problems requires a more complex analysis method, which
is why mathematical modeling has been used to predict glucose in the intensive
care. We have examined mathematical models developed for this purpose, to see
what abilities the glucose sensor needs to have to be able to be used for
glucose regulation in the intensive care when using a mathematical model. There
were several people who read the poster that said that they had been looking
for something just like this work, so there seems to be a need for the research
that we do!

We also had a lecture on the biosensors summer school on mobile diagnostics, titled “Biosensors and systems biology: from sensors to useful insights in biomedicine and clinical practice”. The lecture gave an introduction to how biosensors can play a key role together with systems biology when it comes to investigation of biological problems and in e.g. decision support, and regarding the research in the group it talked, among other things, about the developed multi-level diabetes model.

– Tilda

On Wednesday 18th of May, I had my final bachelor project presentation at ISB Group. I showed the results of my research on prediction of cardiovascular disease in type 2 diabetes patients, using Bayesian networks. Being a Dutch student, this concludes my time in Sweden, that started here: http://blog.isbgroup.eu/#post29 .

I can look back on a productive and instructive 4,5 months at ISB Group. The people in the group are awesome and I’m thankful for including me in all social events right from the start! The group proved to be a place that enables students to thrive at their own research projects and I can definitely recommend it to other international students.

– Gersom

We have been invited to give an oral presentation at the annual systems and synthetic biology conference BioSynSys, which this year is held in Bordeaux, June 27-29. The lecture of Gunnar Cedersund will be held on June 29, and will be on recent conceptual and methodological developments that help us to find more accurate and well-identified predictions and prediction uncertainties, especially in the case of unidentifiability of parameters and single-cell data. Full abstract and title is appended below, and the conference home page is found here.

TITLE AND ABSTRACT

Prediction
uncertainty in the case of unidentifiability and single-cell data – new
concepts and methods

Mathematical
modelling is an integral part of both systems and synthetic biology, because it
can more accurately deal with the complexity of biological data. However, to be
truly useful, the predictions of the model must be in the form of core
predictions, i.e. they must come with a correct uncertainty. In the last few
years, there have been important progress in this field, especially concerning
the important situations of unidentifiable parameters and single-cell data.
This presentation will give an overview of some of these developments.

In the case
of unidentifiable parameters, it has become clear that traditional approaches
based on sensitivity analyses, the Hessian of the cost function, and
sampling-based Monte Carlo approaches all give inaccurate results. In such
situations, one may instead use rediscovered and recently improved alternatives
based on the conditional profile of the likelihood function. Importantly, these
methods can now not only be used for assessing the uncertainty of parameter
values, but for the uncertainty of arbitrary model predictions.

For the
case of single-cell data, problems with unidentifiability are often more
severe: it is often not possible to generate enough data from a single cell,
and averages over many cells provide inaccurate results. In such cases, it is
instead better to use methods from nonlinear mixed-effects modelling (NLME),
which borrows information across the entire cell-population. Using simulated
data where the truth is known, and real data from individual yeast cells, I
will illustrate when, why, and how NLME is advantageous.

All in all,
these new and improved concepts and methods provide important tools for a sound
and correct model-based analysis of single-cell data.

In Linköping, we have a quite rare scientific conference: a conference that has been arranged for over 20 years exclusively by undergraduate students! This conference is called KVIT, and it is arranged by the students in the cognitive sciences programme. The overall focus of the conference is the fascinating mix that is cognitive science: neurophysiology, psychology, IT, artificial intelligence, decision-support, user-interfaces, etc. The specific theme of this year was “Quality of life”, and at this conference we had been invited to give an oral presentation. This presentation was a bit unique because it for the first time outlined some less known long-term plans of our group: to merge and extend our research on a multi-level systems-level understanding of the brain based on mathematical modelling, with fundamental research on the relationship between quantum mechanics and possibilities for free will, and with research on different states of consciousness, such as sleep, narcolepsy, and different types of meditation.

Abstract is appended below, and the entire programme and more information of the conference can be found at the conference home page.

Abstract
One of the big promises of the Information Age is that of systems medicine: that our rapidly growing biomedical datasets will be possible to analyze using
advanced mathematical models, to produce things like automated
diagnoses, personalized treatments, and an improved drug and medical
device development. In this talk, I will go through some recent
developments in this field, to show that this promise is not a
far-fetched, science fiction utopia, but a rapidly approaching reality.
Focusing essentially on my own research, I will show how such
mathematical models now can be used to e.g. replace test animals when
developing new treatments for diabetes, and be used to better unravel
the complexity of the human brain. Using such tools, we can therefore
start to obtain a new type of holistic understanding of the human
organism, into which peces of knowledge both can be examined more
correctly, and subsequently be integrated into a useful picture of the
whole. I will therefore end by a comparison of this increasingly
holistic understanding with such found in e.g. yoga traditions for
thousands of years. What are the similarities and differences, and what
will it take to one day merge such understandings?