Page d'accueil // LCSB // News & E... // Le supercalculateur au service de la lutte contre la COVID-19

Le supercalculateur au service de la lutte contre la COVID-19

twitter linkedin facebook email this page
Publié le vendredi 26 juin 2020

L'installation informatique de haute performance (High-Performance Computing - HPC)  de l’Université du Luxembourg a contribué de manière significative à la lutte contre la pandémie COVID-19 depuis la mi-mars 2020.

Grâce à ses ressources de calcul, le "supercalculateur" et son équipe, dirigée par le professeur Pascal Bouvry et le docteur Sébastien Varrette, ont soutenu les chercheurs de l'Université et les partenaires externes dans plus de sept projets.

Les performances accrues du HPC permettent la résolution de problèmes importants à une vitesse plus élevée. Des tâches qui nécessiteraient normalement plusieurs années de calcul sur un ordinateur de bureau typique peuvent se faire en quelques heures, jours ou semaines sur un système HPC. En ce qui concerne les recherches menées pour lutter contre la pandémie COVID-19, le temps de résolution accéléré est un critère essentiel pour une lutte efficace contre la propagation de la pandémie.

La grande puissance de calcul et les capacités de stockage du supercalculateur de l'Université ont été utilisées pour activer et accélérer la recherche COVID-19 dans les domaines des sciences biomédicales et de la vie, des TIC et des sciences des matériaux. Il facilite entre autres les estimations de la biodisponibilité pulmonaire basées sur l'apprentissage machine et les techniques de modélisation et de simulation des écosystèmes commerciaux pour informer les décideurs économiques au Luxembourg et à l'étranger, tout en permettant de calculer les prévisions futures de la visibilité du virus sur les surfaces.

Les services HPC de l'Université soutiennent quatre projets menés par l'Université et financés par l'appel à propositions accéléré COVID-19 du FNR, un projet de la Task Force COVID-19 de Research Luxembourg et une collaboration entre le Luxembourg Centre for Systems Biomedicine de l'Université du Luxembourg, TU Munich et l'Institut Flatiron.

L'installation informatique de haute performance est un élément de l'infrastructure de recherche numérique et de l'expertise développée par l'Université au cours des dernières années. Elle soutient également l'ambitieuse stratégie numérique de l'Université et en particulier la création d'un centre pour les données et les sciences HPC. Ce centre vise à fournir une infrastructure et des services numériques de haut niveau axés sur l'utilisateur dans le but de favoriser le développement d'activités de collaboration liées à la recherche et à l'enseignement de pointe dans les domaines des sciences informatiques et des sciences des données, y compris le calcul haute performance, l'analyse des données, les applications de données à grande échelle, l'intelligence artificielle et l'apprentissage automatique.

Plus de 1030 tâches ont été programmées sur les réserves dédiées établies par l'équipe du HPC (la plus longue tâche ayant duré 58 jours). Le graphique 1 donne un aperçu de l'utilisation de la charge associée pendant la période la plus critique de la pandémie. L'équipe HPC de l'Université reste déterminée à fournir des ressources et des conseils aux projets actuels et futurs liés à COVID-19.

Graphique 1 : Aperçu des ressources de calcul utilisées par les projets COVID-19 de mi-mars à fin avril.

Le taux élevé d'utilisation des ressources pendant cette période critique montre la forte implication et la collaboration de tous les partenaires de l’Université pour lutter contre cette pandémie. Vous trouverez ci-dessous une liste des principaux projets liés à COVID-19 qui ont fait appel aux ressources informatiques du HPC de l'Université :

Combined In Silico Molecular Docking And In Vitro Experimental Assessment Of Drug Repurposing Candidates For Covid-19 (CovScreen)

Currently, no vaccine or sufficiently validated pharmacological treatment is available for COVID-19. Drug-based strategies to reduce the viral load in patients with severe forms of COVID-19 include the repurposing of existing small molecule compounds that inhibit the activity of key viral proteins or human proteins involved in mediating viral entry or release from the host cell. The project proposes a combined computational and experimental approach to rank an alternative candidate known as drugs, antivirals, and natural compounds, which are commercially available, inexpensive, and safe in humans. The project screens and filters in silico x M~10k compounds using molecular docking and machine learning based lung bioavailability estimations and conducts molecular dynamics simulations for refined binding affinity estimation of the 100 top-ranked compounds. The project aims to provide a fast experimental validation of drug repurposing candidates for COVID-19 from a computational pre-selection of antivirals, drugs, and natural compounds that are inexpensive, have known safety properties, and high predicted bioavailability in the lung. The project is led by Enrico Glaab.

Figure 2: A) Modeled SARS-CoV-2 Spike Glycoprotein overlaid with the SARS-CoV (PDB: 2GHV) unique amino acids are shown. Variable amino acid residue side chains are shown: Green: SARS-CoV Red: SARS-CoV-2. B) Minimised final structure of modeled SARS-CoV-2 spike glycoprotein. Source: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7152904/figure/fig1/

Machine Learning To The Rescue: From Health Recovery To Economic Revival (REBORN)

REBORN is a data science project that focuses on ensuring sustainable economic recovery in the face of COVID-19. The project team applies advanced Machine Learning, business ecosystem modelling (i.e., expert knowledge) and simulation techniques to yield recommendations of economic actions given different scenarios in which the lockdown is relaxed, partially or totally lifted. By interacting with other teams of the Luxembourg Task Force, this project targets high impact for the various sectors of the Luxembourg economy, by providing appropriate data-driven recommendations for political decision-makers. REBORN aims to answer important questions such as: What are the industrial sectors to help in priority? What are the sectors that should be restarted first? What are the possible changes in consumer habits and the impact of neighboring country decisions on commuters? etc. Ultimately, REBORN contributes towards reflections on initiatives to limit the spread of the future economic crisis due to COVID-19 as well as avoid worsening future waves of coronaviruses. The project is led by Jacques Klein.

Figure 3: Tools to strengthen resilience for COVID-19.
Source: https://council.science/current/blog/setting-up-a-data-ecosystem-to-defeat-covid-19/

Privacy Preserving Monitoring of Social Distancing In Public Environments Machine Learning, Computer Vision, Social Distancing, GDPR By Design (PEOPLE)

The project aims to provide a platform to run a comprehensive analysis on the Social Distancing measures decided by the government in the context of the COVID-19 pandemic. It aims to analyse anonymised video data in the city of Luxembourg. The first step is to anonymise the video feed using well-known Artificial Intelligence (AI) models (face blurring). The next step uses other AI models to identify pedestrians and groups of individuals, calculate their relative distances and overall density. Those metrics can then be evaluated over time for different locations and provide valuable insights on the greater or lesser risks of infection spreading based on behaviour. The rules can be used to inform where the police need to focus their efforts on enforcing laws or informing and influencing the public’s actions (or both). The project is led by Raphael Frank.

Virus-surface Interactions In Dynamic Environments (V-SIDE)

The scientific community is trying to establish how long COVID-19 can survive on a given surface (e.g., paper, plastics, glass, and metals). Their properties (smooth v/s rough) play a crucial role during the infection spreading phase. Using the Fourier-transform infrared spectroscopy (FTIR) helps to predict the surface-specific visibility of the coronavirus in the steady-state and dynamic state (e.g., changes in temperature and humidity). Later the obtained model will be mapped into the biophysical model system. Machine learning techniques are used for an optimistic and future prediction of the visibility of the virus on the surfaces. This is the part where the UL HPC software and hardware facilities will be used. The project is led by Anupam Sengupta.

Figure 4: How long does the COVID-19 virus survive on surfaces?
Source: New England Journal of Medicine

Sars-CoV-2 protein structure prediction

Bioinformaticians from Rostlab at TU Munich, the LCSB and the Flatiron institute joined forces to participate in a worldwide effort to predict 3D structures of Sars-CoV-2 proteins. The project was organised by the team behind CASP (Critical Assessment of protein Structure Prediction), who selected ten viral proteins for which no experimental structure is available, nor can homology-based modeling be used to infer structure. The goal is to obtain consensus structures, which would give insight into the molecular mechanisms of the virus, as well as aid vaccine development and the evaluation of possible drug targets.

To this end, the team from TUM and LCSB, helped by collaborators at the Flatiron Institute, trained multiple Deep Learning systems, leveraging evolutionary information from multiple sequence alignments, to predict pairwise distances of amino acids in proteins. The hardest targets of previous CASP competitions were used to assess the reliability of predictions. Computer generated distance maps were then used as constraints to simulate protein folding and obtain 3D structures. Three nodes of the Iris HPC cluster with four Tesla GPUs each (in addition to three similar nodes at the Flatiron Institute and one node from the UCC at TUM) were used to prepare datasets, as well as to train DL models, enabling high quality submissions within the initiative's short-term.

Figure 5: (left) Atomic-level structure of the spike protein of the virus that causes COVID-19 (Source: McLellan Lab, University of Texas at Austin); (right) Visualisations of predicted protein structures from this project.

COVID-19 Task Force

In response to the global COVID-19 pandemic, the COVID-19 Task Force was set up by Research Luxembourg at the early stage of the confinement. As part of Work Package 6 (statistical pandemic projections), researchers from LCSB have developed a computational agent-based model that is essential in understanding the progression of the Covid-19 epidemic in Luxembourg. The statistical projections are required to avoid saturation of the healthcare system and can be used to aid in political decision-making by simulating the epidemic development and its associated outcomes under various future scenarios and strategies.

In order to make informed decisions, hundreds of scenarios with several randomised replicates need to be considered. Parallelisation of the underlying computer code has been key to speed up the simulations and reduce the computing time to an absolute minimum. Thanks to the HPC platform of the University of Luxembourg and its team, it is now possible to launch hundreds of scenarios simultaneously, leading to results being made available to decision-makers within minutes.

Medical image analysis of X-ray and CT of COVID19

During the past weeks, the HPC infrastructure and in particular the GPUs were employed for the AICovIX project. This project focuses on the development of automated solutions for the analysis of medical image analysis of X-ray and CT of COVID19 pneumonia patients using computer vision and deep learning techniques. This work is conducted by researchers of the INS group.

Figure 6: Chest X-rays from a patient with COVID-19 pneumonia, original x-ray (left) and AI-for-pneumonia result (right) (Photo courtesy of UC San Diego Health).