The CLAIRE AI & COVID-19 initiative – Approach, experiences and recommendations – BigStage 2020
12 June 2020 | 13:00-14:30 | Studio The Hague | | |
BigStage 2020 overview
Inspired by successful early use of AI by China, Taiwan, Singapore and South Korea to support the management of the COVID-19 pandemic, on 20th March 2020 CLAIRE, the Confederation of Laboratories for AI Research in Europe (CLAIRE) launched a volunteer effort to help tackle the pandemic. As the World's largest, non-profit network of AI researchers, CLAIRE was quickly able to recruit 150 volunteer AI researchers and to establish 11 research groups. In the talk we will present our experience, the results reached by the bioinformatics group and some recommendations for future initiatives.
- Emanuela Girardi – Founder Pop AI, CLAIRE COVID19 Task Force.
Emanuela Girardi is an AI expert, she founded Pop AI, an association to explain to the people what is AI and how is impacting our lives.
- Giovanni Stilo - Assistant Professor in the Department of Information Engineering, Computer Science and Mathematics at the University of L'Aquila. CLAIRE COVID19 Task Force Bioinformatics Group.
Giovanni Stilo is a researcher in the areas of machine learning and data mining, and specifically temporal mining, social network analysis, network medicine, semantics-aware recommender systems, and anomaly detection. He is founder of the Intelligent Information Mining research group. (https://iim.disim.univaq.it).
- Iarla Kilbane-Dawe - AI and climate change expert. CLAIRE COVID19 Task Force.
Iarla Kilbane-Dawe is an AI expert experienced on how AI can help us reaching the SDGs and specially how AI can help us tackling climate change.
Provided by: Caption First, Inc., P.O. Box 3066, Monument, CO 80132, Phone: +001-719-482-9835, www.captionfirst.com
This text, document, or file is based on live transcription. Communication Access Realtime Translation (CART), captioning, and/or live transcription are provided in order to facilitate communication accessibility and may not be a totally verbatim record of the proceedings. This text, document, or file is not to be distributed or used in any way that may violate copyright law.
>> SANDRA HOFERICHTER: Welcome back to our BigStage session format. The last and the one that’s going to be live, the only one that’s going to be live, it is from Emanuela Girardi who proposed this session dealing with artificial intelligence, timely to our subsequent workshop here in this room which is also about artificial intelligence, and speaker will be Giovanni Stilo and – I see, you are already on line!
>> Hey, everybody! Yes!
>> SANDRA HOFERICHTER: It is very nice to see you! Congratulations that you take the challenge of holding this BigStage live! All others were recorded. I wouldn’t say just recorded, but they were recorded! You are doing it live now.
>> EMANUELA GIRARDI: We decided to try to go live. Yes.
>> SANDRA HOFERICHTER: I wish you great success for your session. I turn the mic and the session over to you. You know that you have to manage your slides by yourself. We don’t.
>> EMANUELA GIRARDI: Perfectly, Giovanni will share them.
>> SANDRA HOFERICHTER: Good luck.
>> EMANUELA GIRARDI: Thank you! I’m the founder of popular artificial intelligence. I’m very honored to be here today to present this initiative, with me, there are two other members of the initiatives and the taskforce created and they are a policy expert that’s joining us live from UK and Giovanni, an assistant professor from the Italy.
First of all, I would like to tell you something about CLAIRE, what is CLAIRE. CLAIRE, it stands for confederation of artificial intelligence research in Europe, it is an organization created by the European AI community and it is the biggest community of researcher, scientists, technologies, experts in the world.
CLAIRE basically is covering all of the different fields of AI and focused on the development of human–centric, trustworthy AI. Today we have 3300 members from 350 research groups, research institutions, and altogether we’re covering about 20,000 AI experts in 34 countries. Basically with such a community of AI experts, when the pandemic started hitting Europe we were like, okay, what can we do really to give support? We offered support of CLAIRE to the governments, health institutions, doctors who handle and to support them in handing the COVID crisis using AI technology. What we did, we launched a call to all of the AI community and we were quickly able to recruit about 150 volunteers, all AI experts and to establish several research groups that were focused on the analysis, mobility data analysis, on bio formatics on medical digital analytics, robotics, last but not least, scheduling resources and management. Basically this group, they started working together and they were focusing on two things, on one side they were focusing on collecting and curating resources, basically datasets. We heard from the previous presentation, it was that data availability, it was one of the main issues. One of the main focus of our group was focused on that.
On the other side, basically they started developing new projects in several AI application fields really to give a concrete support to government, health institutions and doctors. What we did, we already got very interesting outcomes from all these research groups. In particular actually we got some impressive results from the bio researchers. What they did, they assembled an impressive repository of datasets, resources to help scientists and doctors in reporting activities for fighting COVID‑19. Today basically it will be presented on how the group organized the work, the use, and which of the main results that they already achieved in the group and they’re still working on these actually.
Then after Giovanni, we have another sharing with what we learned with this experience. This was a very interesting experience, managing COVID‑19 taskforce and we learned a lot. So we would like to share which are our recommendations for further initiatives like this one.
Okay. Thank you.
Now I give the floor to Giovanni presenting the results of the bioformatic research group.
>> Hello. I start by sharing the screen..
You’re able to see it.
What I will present today, it is the results of the CLAIRE COVID taskforce in the field of bioinformatics. As already remarked, the aim of the taskforce is to work with AI techniques and more specifically we are working in the field of clinical and life sciences. We bring together some experts from the field of studies, plus artificial intelligence. More precisely, what we would like to do, we somehow would characterize the disease and more specifically the COVID and starting the interaction of the virus with the human cost and hopefully filtering, retrieving, generating new drugs that can be adopted, they can be employed to fight the COVID.
At the end, we would like to understand and predict some genetic features of the virus.
The taskforce, it is composed by 18 active members that are going to interact across Europe and their skills are mainly focused on the field of artificial intelligence, machine learning, data science. More specifically, they have experience with regulatory and network data analysis, data analysis and sequencing and variance analysis. The way that we have worked, we have started to work together at the beginning of March, it was to somehow crowdsourcing the needs of the physicians that we’re in contact within a collaborative way and to develop some tasks that we would like to explore and if we need some specific expertise, we try to bring them inside the collaboration. More specifically, we have started to work on three tasks, and the first one, it is the ground task that we have started to develop, it is in the field of drug repurposing problem, but we would like to also explore some predicting models to understand if a specific patient needs to be recovered in an intensive care unit or not. At the end, we would like to also develop artificial intelligence techniques to promote the screening of virus for individuals.
Let’s see which is the first problem that we have. It is called the drug repurposing problem, in other words, consider that you have millions of tested approved and readily available drugs, how can you decide in a very short, she small amount of time which one could be the more effective to treat a new virus? You don’t have time to test everything. This is where the artificial intelligence can help out.
So our way of approaching this problem, it is to first collect some specific data such as the protein‑protein network, the human‑viral interactome and a list of candidate drugs and we’ll apply the graph network, machine learning and network data analysis techniques and we divide this problem in subtasks we’ll explore. The first is to prepare a candidate interactome and we create an encoding of all of the information that then next we’ll use to train a specific way to allow us to assess at the end the set of drugs that we would like to explore.
So to be more specific, as already said by Emanuela Girardi, we have achieved to bring together a lot of information from the field. The aim of this subtask, it is not only a functional for our official exploring, but we have realized that if someone wants to start to do research in these specific subtopics, meaning to select the results, accessing them, this is a lot of time demand and task. Honestly, it is taking a lot of time to do that. For this reason, we decided to collect them by selecting them and documenting them and we have all of this information in one single repository that is very fast to be accessed.
Let’s see which are the information most specifically we have collected in these.
The first, most important information, it is the interact movement. The collection of that interaction with the human body, and on top of this, we have collected some other information such as the domains that basically are going to classify the probing that are responsible to produce some specific function and we also collect with the families, the families, they’re basically a group of – of proteins that explore similarities in their sequence or their structure and show pathways that are a series of molecular events that will produce that biological and on top of them, we also explore the genome that it is very specific, that it is dedicated to the biological process, the cellular component, and also we collect the interaction, in other words it is the least of possible drugs that you can provide and we also collect the data structures, so there is a structure at the clinical level and it seems if you’re going toed a men center several drugs to the patient, they could be affected. Another piece of information that you need to put in place so that you can understand the interaction along with the drugs. To finish all of this information that we have seen now, we have also collected the disease information, the general disease information that’s not virus that normally are going to attack, to change the behavior of the disease and at the end we collect the topmost important virus that can be in the human biological system. So these are all in one system, one dataset repository, but it is by scientists, others, physicians, they can be ready to start working on the problem, not necessarily on the data. Other things that we have done while we are assembling all of this information together, it is, okay, we would like to select all these results from several public accessible databases and so on. We would like to select them without narrowing too much on the problem, the day‑to‑day problem, so you will find information about COVID and much more information that can be used in the future, not only to solve the drug processing programme that was our first aim, the aim that we have in mind when we start working on this results.
Just to have an idea of the power of the results, we can see how it looks like, a schematic representation of the human interactome, this is one proton, and we have one line, if the two are going to interact, one with each other from the biological point of view, and in this image, we have highlighted with orange and this portion of the subset of all of the possible protons because of the positive protons are thousands, it is not possible to represent them clearly. We limited this only to the enabled protons affected by the COVID, the data represented in orange in this picture.
Just to finish, and to give some hints, basically the repository, it is distributed publicly using the GitHub platform and you can directly access that repository with the data and the description using the link below or otherwise using the code on the left part of the screen. If you are interested on our efforts to describe the process which we have selected from which original source of information, we have taken them, you can look at the description and so this contains the description and this description can be accessed and using the code on the right part of the screen or otherwise you can find it also in the repository. I guess that’s all.
I give the microphone to you. Thank you, everybody.
>> Despite that challenge, despite other challenges such as the access to datasets, to the lack of open licenses, standards for such datasets, so on, some effective work was delivered, including in very large, complex teams. I think that the largest team that we had working on a particular research topic had 49 researchers working underway at any given time. That obviously led to a lot of diversity of ideas and a lot of diversity of concepts. Actually, what we found, it was unsurprisingly again the larger the team, the longer it took people to focus on a few key ideas to work on. Nonetheless, the volunteers showed that in a matter of a few weeks, amounting to just over two months, they were able to deliver some really successful, interesting products such as what was mentioned.
There is definitely more forethought required allowing for the fact that crisis of this type, it is unavoidable. Some keep crisis of these, it may not – an observation out of our experience, it is to build a bit more domain expertise in the potential future crisis while this work obviously goes on all the time by research groups everywhere, there are areas that we could consider focusing on, to prepare for future crisis, whether pandemics or not, they’re well documented, well cataloged and should be the work of future research to build relevant domain expertise.
Over and above that, what we would say, it has been a very interesting experience to not only deliver work with very large groups, often using quite basic communication tools, but to see those then move on into much larger and more permanent activities such as establishing large complex research proposal such as the work to provide high performance computing support that’s been offered by several different sources for work on COVID‑19. It is interesting to see what started out as a volunteer effort often without the direct support of authorities or domain experts have evolved very quickly into live proposals and ongoing activities that will allow people to deepen their expertise, to continue to provide research in these areas in all likelihood for years to come.
Overall, although it started out as a volunteer effort, it has moved quite quickly through the efforts of the volunteers and through providing a few bits of communication facilitated by the taskforce itself into substantial pieces of work and serious and substantial research proposals for the future.
Thank you very much.
>> EMANUELA GIRARDI: Thank you. I just am going to close just saying, noticing something, that what we learned actually from this experience on top of everything in the recommendation that Iarl just shared with you, Europe needs, the one in the white paper for AI, released by the European Commission last February, it was the lighthouse for AI. This is something such as CLAIRE, it needs really a point of reference for AI development and activation and especially situation like the crisis that we just lived because you need a point of reference, a central point of reference that you can get in touch with government, institution, you can get in touch with and that can really help supporting application, technology, devices, and everything because AI technologies, artificial intelligence, it can really help us managing this pandemic and also can give us a support to the whole community and society.
This is the key learning link for me. I thank you, of course, and I want to thank you EuroDIG for inviting us and for giving us the opportunity to present you CLAIRE and the CLAIRE AI and COVID taskforce.
Thank you. If you want to contribute, get in touch for further information, please do not hesitate, we would be very happy to get in touch with you. Thank you.
>> SANDRA HOFERICHTER: I think this worked extremely well. Congratulations.
I think the next session that starts at 2:30, it a good follow‑up, it speaks about fighting COVID‑19 with artificial intelligence, how to deploy solutions we trust.
So for now, I give over to our relaxing music for another half an hour. Get a coffee, be ready here at 2:30 sharp. We’ll start with the next workshop in all three studios.
See you then.