 So some of these are in the speaker and you'll talk about it. Okay, so first of all I would like to send the organizers to pulling this together, I think this is a very important topic. So I will talk about this NMR project, which is touching many of the things we have discussed, but now today I will talk about the point of view of first-produced ability. But before that I will shortly tell what this is about, because it's important to understand what it is. So this started when I was positive in Lund. I was working with the Solid-State NMR Experts, made with Jack Farrere and Daniel Topkart. And they had this fancy new pulse sequence, which you can use to measure order parameters of limited bilayer using carbon-13 natural abundance. So before you need a label to put this and so on. So what is the order parameter for the limited bilayer? So for each hydrocarbon segment that's a little bit you can measure this number, which is called order parameter. So this is the angle of the CH bond with respect to the membrane normal. And this is the ensemble average. So this is the average over the conformation is liquid molecule sampling. It's well-known that if you measure these order parameters for the actual change you calculate from every simulation you get really good agreement, which is for many people this is well-known, but for the animal people this is actually quite common, so this is very quick. It means that the conformation of the actual change are correctly sampled. I wasn't very excited on this, but then when I was there I realized something that you can actually measure these order parameters also pretty clear of backbone and the head group, which I didn't realize before. So then I compared those to the experiments of the burger model at the time and the agreement was not that good, which means that in this model the conformation is sampled by the clear of backbone and the cola and we're not correct. I think even more important is that it turns out that the order parameters of the head group, alpha and beta carbon proportionally change when you add charges to the bilayer. So there's a linear relation between the amount of charge in the bilayer and this order parameter, which means that you can measure ion binding affinity using these order parameters because the order parameters depends on how much ions you have bound. The good thing is that we can directly compare these to the experiment so we can directly compare the ion binding affinity and between simulations in animal. So when I did this comparison, the burger model, it was not very good. So the figure is not very clear, but for me it was enough to say that the binding is too strong. So the cations are binding too strong. Okay, so then I, okay, what I learned. I learned that this burger model is not very good for the head group, this ion binding is not good. So what can I do? This, at the time, this was very widely used model. So what can I do? I could write a publication perfectly doable, problem is that nobody would probably care because this is not a real problem, this is just a negative result. Which is fine, but I don't think it pushes much. So the other option would be that I will take all the force fields we have. I will try to fix this, I will try to understand how this works. So this was too much work. So at the time, 2013, we had a lot of these discussions with Margot Smith and with another guy. How we should actually do science and there were people saying, oh, I'm fine, then I thought, okay, let's just do it. So we tried to do why to do things. So how we started, I wrote this kind of manuscript telling what I had observed, we published it in the archive. Then we launched this open collaboration. So this diagram comes from Polymath Project and Linux collaboration and stuff like that. So we basically started the blog. We presented an open invitation to the field to contribute, give data, ideas, whatever you do. So everything is done publicly. So all the contributions are done publicly. If you contribute, you will be offered an ownership when we publish. The ownership is based on the final self-assessment. So if you think you should be there, you will be there. Yeah. So, okay, one thing I want to find out is there a little bit of misconception of what we are trying to do. We are trying to find any simulation models of lipids which currently describe biological mineral and membrane properties. We are not comparing force fields. We are not building a data bank as a first goal. These are side products which may be even more important than this. But this is the main goal. Because some people think that force fields comprising people. Okay, so we started 2013. We have now 42 contributors who have contributed. We have published three papers. We have three manuscripts. These are the papers. These are the people who have contributed. I will quickly show. I will try to go fast because... So what do we get out from this kind of thing? We have actually a lot of data research, just like very small fractions of the examples that we can learn. So these are the head group order parameters for POPC. Now you should sit here with the blue-shaded regions. You should see the experimental data. The simulation results from the different colors. So they are pretty much all over the place. This is part of PES. These are the structures of the PES lipids from the different force fields. The data is a bit overwhelming over here. So we have also done this kind of quick... We have a metric of which force fields. In terms of the structure of the head groups, we can get this kind of... If you're thinking about what model you should use, this could be useful. The ion binding was a very important thing. I don't have time to go into the details. The message is that the cations are binding too strongly in all simulation models available. Sodium and calcium are binding too strongly to lipid bilers. This is the best part. So you can see that all the simulation models go like this. And this is that for me. Cations are binding too strongly. I don't have time for this, but I will show... We can also fix this. So we have also a new model, which gives a good result because of the implicit input of the electronic part. But I just show this that we are also making actual progress. I was asked to talk about the reproducible, so I will go there now. I was asked to give practical examples where this has been a problem. We are looking at force fields, which would actually work. What happens is we take force fields by other people. We try it before. When we try this, we should actually make it correct. We should be doing it right. So then what happens is that if you do it, you check if you got the same results as people before. If you don't, you have problems if you do it. So there are three types of problems, general problems. One is that there are some force fields which is done for some package where you try to use the other one and then maybe there is a problem, maybe there is another confusion. One problem is that others do the parameters. They put them online but then they are not the same what they actually use for one reason or another. Or then they put the parameters online and they just change them without telling anyone, which is confusing other people. So I will have some practical examples of this. And then I will quickly say what I think we should do. So this is a classic chart. As we will change the order of parameters of chart 3640. So this is what they show in the paper. Very nice agreement with the experiment. But then no matter what you do, no matter what I have seen people doing, you always get bigger than in the experiment. So here is the longest experiment. These are from different simulation papers from different people, maybe a little bit different conditions. But they are pretty much in line with each other. Not with the parameters, but not with the experiments. The all version of chromax is a different thing, but with the newer one. Consequently, I cannot, like people said, that charm gives you a good picture, a good model of liquid bilayer, but maybe in the original paper it is, but I cannot do it myself. So it's a bit annoying. Another example is, or what is, or what is, with the amber, amber limit. So this amber, converting amber plus the chromax, it's been mentioned a few times, and it's known to be a little bit annoying. So what happens was that there was this model, there was this, at the time instead of the R amber limit model called goth limits. So then we transfer it into the products, we run the simulation, we get the area per limit of 61.6. In the paper they report 63.9. So it was clearly more condensed. We did quite a bit of work to try to get this right, we never managed. So then we concluded that, yeah, okay, it's the amber conversion problem. But then the same guys came up with the new model, and we did exactly the same thing. And then we got 64.5, then we got 64.6. So some of them we were able to do it. I don't know, I think actually probably there was something wrong with the parameters delivered with this one, but it's, obviously it's not the, instead of the R model anyway, but we lost quite a bit of time for this kind of thing. So this is an example of this tool, or either one I don't even know. This is a little bit different, so this is a clear example where others share parameters online, which might not be good. So there are OPLS parameters, we call them macro, best on the first names of the first others. So they have parameters, for example, for P O P S available from this address, which we were using. So this guy, Thomas Biggs, I don't know, an Amnesty relation guy. Here we're all doing the block. So he took the parameters, he downloaded the parameters and took a little bit of a look at them. Well, they noticed that, so this is a real P O P S molecule. We have a double bond in the SN2 chain. So he realized that there's double bond in the SN1 chain in these parameters. So it's a different molecule. So they say that it's P O P S, but it's actually all P P S. So they have the acetyl chains in the wrong positions in the closure of the backbone. Well, then he said that, okay, I will just fix it, then we will do it again. And then he did it. And then we had the P O P S. But then I was looking at the figure, I was writing the paper, and I was looking at the figures. So then I noticed that if you look at this, these are cell line structures. There's only a couple of forces. There's a bit more. But there's something different here. After looking at the while, you realize that this is a different sterile isomer than the other models. So this is the L isomer. This is the D isomer. And this is what you have in the nature. So it turned out that actually in this file source, they had the initial structure which had the wrong sterile isomer. So then we fixed that one, but then there started to become more and more problems. And the ion binding thing. So we did ion binding. We looked at the ion binding of anything. Against the NMR data. So we did this for TROM 36 from TROM going through 2015. For the NMR, it's two paper. We get way too much deep breath in the order parameters. Here's all the calcium binding. There's the calcium density in the water. So in 2015, with charngui, all the calcium is binding to the membrane. For a number of reasons, we did this again in 2018. We get the red thing. Which is much less binding we see there than as the profile as well. Now there's almost zero binding. So what happened between 2015 and 2019? We gave them out to the charngui people and they told that the alloy added this NBFX parameter from Ben Oro in the primary because calcium was binding too much. Which is fine. The problem is that they never told this to anyone. Okay, solution. We all know this. There's nothing new. We all know that there are these kind of problems. You know, it's sort of... Yeah. There's nothing to do with this. So I think Ben Oro's solution is that we should have all the data, all the raw data, all the steps from raw data to the actual manuscript publicly available. Which means that you could perfectly track the path from the raw data to the actual manuscript how this can be done. So Ben Oro's solution is to put all the things in the GitHub. We have predefined GitHub folder repository for the manuscript so the idea is that we have one file where we always have the manuscript that is in text form which is reading figures where the data is presented. It's reading figures from the folder of figures so you know that you find the figures from here. This folder is reading the data which is derived from the raw data which is in a certain folder. And then we have the scripts which is deriving the data from the raw data. So then we would have predefined manuscript folder structure. Everybody would know that you can find all the steps from here. Raw data is more difficult because files are bigger so we can put it in the GitHub. At least. So what we do is that we put it in Zeno. Now when this is reading the data it downloads from Zeno to Zeno. So there are Zeno to wins. So it downloads the data and analyzes it pushes there. So the data is not actually here you never push the raw data because it's too big. So all the data is in Zeno. And then it's in the tables. In the papers for example. There's a citation to each the citation takes you to Zeno. Oh sorry. I took a bad example. This picture. But it doesn't matter. It's somewhere. And then we have this. I have this I mentioned before. So then we have this index database which is also indexing this data. So there are these limits. So you can search the simulations of these limits in different temperatures. And then you have a link you can go there and try. This should take you to the GitHub repository which links to the raw data. And there's a script which downloads and then does the analysis. I would like to talk about this a bit more but this is not reproduced in this one session. So I wrote I wrote a short description how this works in the Google documents. So you can go there and read it if you're interested. Okay. Conclusions of this open collaboration is good. It's possible to publish the whole progress of reproducing the of doing publications which improves whatever you want to call it. I call it reproducible but it's actually trying it out. And we have this good database. We have anyone with its workshop coming in May in Berlin. So if you're interested in the open collaboration this is completely done with Martin Metten and with the other they are these are the finding sources and there are this kind of problem it was said before that you might get, I'm employed un-activated stuff like that. I was like that a couple years ago. But then Paul Jungworth was able to convince that I should keep on doing this and he covered me very much. I pretended to work in Prague before I got the academic bonus. No, I have funding for 5 years. Okay. There's nothing to make of this battle. I agree with this one. Thank you.