WEBVTT 00:00.000 --> 00:13.000 Hello, to everyone, first I would like to thank the guys for giving me the opportunity to talk in this excellent conference. 00:13.000 --> 00:23.000 My name is Jose Spinozacarrasco, I'm working in the LAP of Notre Dame's LAP at the CERG in Barcelona. This is the LAP where NEXO was originally developed. 00:23.000 --> 00:36.000 I'm also an NFC or team member, and today I'm going to introduce you an NFC or platform, which is a standardized pipeline for HAP for Plotting Statue of Nature Methods. 00:36.000 --> 00:49.000 So maybe if you don't know already, you may be wondering why it's interesting to predict Plotting Statue and the thing is that once you have the Statue of Plotting you can infer its function. 00:49.000 --> 01:03.000 And once we know the function of this Plotting, we can do some applications, like for instance, the Stabilized Statue for Biatenological Applications, so that they work in higher temperatures with higher performance. 01:03.000 --> 01:24.000 We can do things like design, for acting in its active side, in a intelligent way, or we can do more basic signs and we can study the evolution of this given Plotting, comparing its structure with other Plotting families and species. 01:24.000 --> 01:39.000 And it's possible to get the Statue of Plotting using a experimental method, but the thing is that as you may imagine, this is very time consuming and also very expensive. 01:39.000 --> 01:49.000 And that's why it has been a long-standing program in computational structure biology, how to predict structures using computational methods. 01:49.000 --> 02:05.000 So that there's even a competition, which is called the CAS contest, and during this contest, the Stabilized Statue of Plotting families gather and they compare their methods, and one of them, up to the other, and it wins in different data sets. 02:05.000 --> 02:24.000 And this is in the framework of this contest, it was when, for the first time, a photo show that AI-based methods can outperform existing methods and not only that, but they can, alpha photo was able to achieve almost experimental accuracy. 02:24.000 --> 02:52.000 So this was a big breakthrough in biology, and it was awarded the 2024 chemist Nobel Prize, as you can see here, but more in practical terms, what I'm showing you here in this ball or in this circle is how many structures were available in the PDV database, which is the database where people were were the position, the structures of the Plottings that were played at the experimentally. 02:52.000 --> 03:01.000 And after the alpha photo, inception, you can see how the number of structures a bio has employed. 03:01.000 --> 03:11.000 And indeed, also the number of tools that can predict structures using a-based methods has employed. 03:11.000 --> 03:20.000 Here, you have several of them, and we have mainly two categories, most of them, as you can see, are based in evolutionary research, DBM models. 03:20.000 --> 03:29.000 And this means that these tools, what they do is that they search for similar sequences in databases, and then they perform multiple sequence alignment. 03:29.000 --> 03:38.000 And with these multiple sequence alignment, what they want to achieve is to know which parts of the protein of the sequence of the protein are evolutionary concerns. 03:38.000 --> 03:49.000 These parts of the protein are important for the function, and they tend to be maintained in evolution for being absorbed at the protein can still have the function. 03:49.000 --> 04:01.000 And after this, with this information, and with this sequence, this is applied to an neural network, which is the one that infers the structure, and after several steps of refinement. 04:01.000 --> 04:18.000 We have for their tools that use large language models, like ESM-4, and in this case, these tools are faster, but sometimes dispenses of accuracy, mainly when dealing with long sequences. 04:18.000 --> 04:25.000 So, as always, in competition, there is a trade-off between speed, accuracy, and those cost. 04:25.000 --> 04:35.000 Yes, I wanted to say something else. I got lost with it, I mean, sorry. 04:35.000 --> 04:48.000 And of course, users, depending on what they want to do, they will use one of the tools, or they will like to use several of these tools, and so on and so forth. 04:48.000 --> 04:59.000 But actually running these tools is not to stay forward, because they try to provide you the images to make it done, but even though it's hard. 04:59.000 --> 05:23.000 And why it's hard, because all of these tools, they rely in several libraries, different versions of the libraries, so you have to make all them available in your environment, and also the tools that are based, that need the MSA step, they also rely in very big databases that you have to have in your environment. 05:23.000 --> 05:41.000 And this is why we implemented NF Core Plotting Fold, here you have the typical NF Core Metal Map, and you can see here in this first, to go through that this pipeline allows you to download all the databases that you need for these tools. 05:41.000 --> 05:51.000 And this is a one-shot workflow, this means that you don't need to run it each time that you run the pipeline, you run it once, and then you can provide it to the pipeline with the human parameters databases. 05:51.000 --> 06:00.000 And then you can see the alternative paths that you can follow in this pipeline, this is the development version, by the way. 06:00.000 --> 06:12.000 So here on the top, you have tools that are based in evolutionary driven tools. 06:12.000 --> 06:23.000 In this case, these tools perform both the MSA step and the prediction step in the same process, so you can not split them. 06:23.000 --> 06:40.000 Then here you have these two other paths, the one in green, and the one in yellow, the green, it's alpha-fold 2, alpha-fold 2 doesn't allow to do it natively, this separation between the MSA step and the prediction step, but we have modified to be able to do it. 06:40.000 --> 06:48.000 And also, call up for MSA, call up for Lamboz, that in this case this is native, and we just take advantage of it. 06:48.000 --> 06:54.000 And here you have ESM Fold, which is the one that it's using the large language model approach. 06:54.000 --> 07:08.000 So the pipeline is giving you the PDVs, the structures of the proteins, but within this part is also very interesting, this is something that we have been working on is that for each tool you get a report with all the information of the accuracy. 07:08.000 --> 07:18.000 But also you also get another report which allows you to compare the performance of the different tools that you have used in your run. 07:18.000 --> 07:28.000 And here I am just summarizing some of the futures of these pipelines as I show before it allows you to plan on load the databases and the parameters of the models. 07:28.000 --> 07:36.000 Everything is containerized, so it means that you don't need to install any of these tools or its dependencies. 07:36.000 --> 07:50.000 It's easy to configure, and this is important for these tools that allow you to run the prediction step, separated from the MSA step, because MSA step does not get any advantage of being run in CPUs. 07:50.000 --> 07:53.000 This way you can configure it to run in CPU and CPU. 07:53.000 --> 08:05.000 And this is something that it's actually, can be done thanks to NXO because NXO separates the configuration from the logic of the pipeline. 08:05.000 --> 08:10.000 So it's just need to provide a configuration file and you get it. 08:10.000 --> 08:17.000 And as I said, as I mentioned before, it allows you to do methods of benchmarking. 08:17.000 --> 08:23.000 And actually this is also interesting because this is something that we are pushing in our lab is half pipelines. 08:23.000 --> 08:31.000 This means pipelines that perform that where you have several tools that perform the same thing. 08:31.000 --> 08:43.000 This will allow users to compare these tools among them, but also the developers of the tools or people that are interested in comparing the performance of these tools to it using these pipelines. 08:43.000 --> 08:52.000 It have two reporting capabilities that I already mentioned, and it's part of NF Core which have several embedded as I will show you. 08:52.000 --> 08:56.000 Here in this slide, I'm showing the comparison report. 08:56.000 --> 09:01.000 Each of the colors depends one of the tools that we have used to predict this structure. 09:01.000 --> 09:04.000 This is a sequence that we use for the testing of the pipeline. 09:04.000 --> 09:08.000 So it's a single sequence and several tools to get the structures. 09:08.000 --> 09:10.000 You can see how you can spin them. 09:10.000 --> 09:12.000 You can choose which one you want to show. 09:12.000 --> 09:17.000 Then you can see here the accuracy of each of the tools. 09:17.000 --> 09:22.000 The average accuracy. 09:22.000 --> 09:33.000 What you can see afterwards is that the accuracy along the sequence. 09:33.000 --> 09:38.000 And then finally, you can see the coverage along the sequence as well. 09:38.000 --> 09:43.000 Let's see if I... 09:44.000 --> 09:48.000 So... 09:48.000 --> 09:51.000 Yes. 09:51.000 --> 09:52.000 Yes. 09:52.000 --> 09:54.000 I will stay in this one. 09:54.000 --> 09:55.000 Ah. 09:55.000 --> 09:56.000 Is it opposite? 09:56.000 --> 09:57.000 Okay. 09:57.000 --> 09:59.000 What is that? 09:59.000 --> 10:02.000 Sorry about that. 10:02.000 --> 10:08.000 Okay. 10:08.000 --> 10:14.000 So this is a real application of the pipeline that has been done by Kiran Roewell in Sydney. 10:14.000 --> 10:24.000 What he has done is that he wanted to prove these hypotheses, whether a sales evolved via San Bios is between 10:24.000 --> 10:27.000 two studies in bacteria and in the same producer in Arcaya. 10:27.000 --> 10:29.000 So he got all the... 10:29.000 --> 10:34.000 He's... they sequenced all the genome of one of these Arcaya. 10:34.000 --> 10:40.000 They then use... in a corporate info to predict all the structures of these... of these Arcaya. 10:40.000 --> 10:45.000 And for this, he was using alternative methods because for some of them it was easy. 10:45.000 --> 10:47.000 He was using just ESM fault. 10:47.000 --> 10:48.000 And that's all. 10:48.000 --> 10:53.000 But when he was dealing with more long sequences, sometimes it was not working. 10:53.000 --> 10:56.000 Then he moved to Alphafall 2 or in some ways that weren't even longer. 10:56.000 --> 11:00.000 He used or more difficult to infer he used Alphafall 3. 11:00.000 --> 11:05.000 And of course it was very interesting having these reports to compare these performance. 11:05.000 --> 11:13.000 And then he used Faultsick to see whether these structures actually fall close together with the bacterial sequences. 11:13.000 --> 11:18.000 But this is... I think I'm a little bit out of time. 11:18.000 --> 11:21.000 You can beat the paper here. 11:22.000 --> 11:24.000 This is not the last version of the... Sorry. 11:24.000 --> 11:26.000 Well, I will move here. 11:26.000 --> 11:31.000 And another interesting thing of NF Core is that... 11:31.000 --> 11:37.000 NF Core makes that the pipelines can last longer. 11:37.000 --> 11:39.000 And why is this? 11:39.000 --> 11:41.000 Because sometimes you work in your lab. 11:41.000 --> 11:42.000 You have a very nice pipeline. 11:42.000 --> 11:44.000 But at some point in the funding stops. 11:44.000 --> 11:48.000 Or maybe you don't have more in-setific interest on this pipeline. 11:48.000 --> 11:49.000 But this pipeline is a very nice pipeline. 11:49.000 --> 11:51.000 And it's useful for a community. 11:51.000 --> 11:55.000 And what happens in NF Core is that as you can see with the sample of this... 11:55.000 --> 11:59.000 This smaller-nastic pipeline is that the people that have started the pipeline 11:59.000 --> 12:01.000 is the life-love at some point. 12:01.000 --> 12:03.000 They even stop working on it. 12:03.000 --> 12:06.000 But all the groups were from... 12:06.000 --> 12:08.000 Both from industry and from... 12:08.000 --> 12:11.000 Academia were already contributing to it. 12:11.000 --> 12:12.000 And they just took over. 12:12.000 --> 12:14.000 So this means that you... 12:14.000 --> 12:16.000 Did that you pipeline last... 12:17.000 --> 12:18.000 Not forever. 12:18.000 --> 12:20.000 I would say that longer than if you are in... 12:20.000 --> 12:22.000 Isolated in your isolated lab. 12:22.000 --> 12:24.000 And here you have all the contributors. 12:24.000 --> 12:25.000 That have contributed to the NF Core. 12:25.000 --> 12:26.000 But in full pipeline. 12:26.000 --> 12:28.000 From the Sierra Lea in Barcelona. 12:28.000 --> 12:29.000 Also UPF in Barcelona. 12:29.000 --> 12:33.000 We are technically working with people in Sydney. 12:33.000 --> 12:34.000 In the UN's. 12:34.000 --> 12:35.000 The UN's. 12:35.000 --> 12:36.000 And also from Australia. 12:36.000 --> 12:37.000 Come on. 12:37.000 --> 12:38.000 And then your university. 12:38.000 --> 12:40.000 University of Korea. 12:40.000 --> 12:43.000 So if you want to join us, here you have... 12:43.000 --> 12:44.000 S7 out of the ways. 12:44.000 --> 12:45.000 How you can do it. 12:45.000 --> 12:47.000 So you can join the NF Core. 12:47.000 --> 12:49.000 There are several ways to do it. 12:49.000 --> 12:53.000 You can also join a Slack, which is the main communication channel of the NF Core. 12:53.000 --> 12:58.000 And there we have the protein-full channel, which is for the users of the pipeline. 12:58.000 --> 13:01.000 And the protein-full web channel, which is for the developers of the pipeline. 13:01.000 --> 13:03.000 And you are very welcome to come here. 13:03.000 --> 13:05.000 And also the GitHub repository. 13:05.000 --> 13:09.000 We have a paper in preparation that we want to submit soon. 13:10.000 --> 13:19.000 More importantly, we have all the tools that have shown you are not yet in the release version of the pipeline. 13:19.000 --> 13:20.000 They are still in development. 13:20.000 --> 13:25.000 But we are planning this release for, I think, in two weeks should be ready. 13:25.000 --> 13:27.000 I wanted to be ready for this presentation. 13:27.000 --> 13:28.000 But it was not. 13:28.000 --> 13:32.000 And here you have other things that we want to optimize on the pipeline. 13:32.000 --> 13:34.000 But you can take a look. 13:34.000 --> 13:36.000 And maybe we can discuss afterwards. 13:36.000 --> 13:40.000 And finally, I just want to finish thinking my supervisors, 13:40.000 --> 13:42.000 I think Notre Dame, of the people, of my lab, 13:42.000 --> 13:47.000 especially Athanasios Vatsis, who was the original developer of the pipeline. 13:47.000 --> 13:51.000 Also the people in Australia, marketing from Korea. 13:51.000 --> 13:56.000 The people from Zekera, both the NF Core and the XO community. 13:56.000 --> 13:59.000 And also the AWS. 13:59.000 --> 14:01.000 Open data-sports and simpler. 14:01.000 --> 14:05.000 Who is allowing us to host all the data-basis in, in, in, in, 14:05.000 --> 14:06.000 in, in, in, in, in, in, in, in, in, in, in, in, in, in, in, in, in. 14:06.000 --> 14:08.000 So thank you very much. 14:08.000 --> 14:10.000 And I am open to questions. 14:10.000 --> 14:12.000 Thank you very much.