ExaScience: HPC software lab
Use high-performance computing (HPC) to scale your software: decrease runtimes, cut compute costs and enable new applications without hardware investments.
Doing more with less. When it comes to computing, that’s the challenge we face today. We need to crunch larger amounts of data in shorter amounts of time – without building acres of server farms that run on terawatts of energy. We want to move applications from the cloud to our desktop – or even to our smartphone. Or we crave broader insights but lack the necessary data.
High-performance computing software presents one way to tackle these problems of scalability: it aims to get more out of available hardware and data.
Imec’s Exascience lab leverages its background in data-analytics and AI to develop next-generation HPC software for every level of the computing stack:
- supercomputers and cloud data centers
- individual computer systems
- embedded devices or computer boards
HPC software: from life sciences to Industry 4.0
Imec’s ExaScience lab has an exceptional track record in the development of HPC software solutions for the pharmaceutical and healthcare sectors. The combination of ExaScience’s software knowhow and imec’s expertise in domains such as connected health and life sciences enables us to efficiently partner with bio-informaticians and clinicians to scale their applications. We complement existing teams with our focus on performance engineering and/or privacy-preserving AI.
Explore this article for more insights into the value of HPC software for life sciences.
One way in which high-performance computing benefits life sciences is performance scaling of big-data analytics. For example:
- high-content image analysis: feature extraction time of 45 TB of images from months to weeks
- genomics: whole genome sequencing from days to minutes
- chemogenomics: machine learning from weeks to days
- medical statistics: unsupervised generation of patient trajectories with temporal clustering in hours instead of weeks
Clinical and pharmaceutical research also benefit from the scaling of available data through privacy-preserving AI. This allows different parties to derive insights from each other’s data, without sharing any data or models.
Imec’s ExaScience lab currently brings its HPC software expertise to use cases outside of the life sciences sector. One of those fields is Industry 4.0, where there’s a growing need to quickly mine massive amounts of data for actionable insights. Or where privacy-preserving AI can be the key to optimize a complete production cycle by tapping into machine data from different vendors.
HPC software techniques
How do we manage to get more out of the available computing hardware? We use techniques such as:
- high-performance machine learning software (deep learning, Bayesian probabilistic matrix factorization, unsupervised clustering, ...)
- concurrent software engineering
- distributed computing
To achieve our goals, we combine our wide software knowhow with an in-depth understanding of hardware architectures – from cloud infrastructures to embedded computing boards, leveraging compute elements such CPUs, GPUs, FPGAs, TPUs, or specific interconnects like Infiniband (when applicable).
HPC open-source software downloads
Several of our HPC projects are available as open-source software tools:
- elPrep is a DNA sequence analysis software solution that’s up to sixteen times faster than other programs. Download on GitHub
- SMURFF is a high-performance framework for matrix factorization that can be used for recommender systems for drug discovery, genome editing, ... Download on GitHub
- Pargo is a library that provides convenience functionality to turn otherwise sequential algorithms into efficient parallel algorithms for the Go language, complementing Go’s excellent support for concurring programming. Download on GitHub
Work with us
Do you think HPC software could help you with scaling challenges? Then we’re happy to assist you in all stages of your development, complementing your team with our performance or privacy-preserving AI experts:
- feasibility study – We investigate the viability of your project and, if possible, define its specifications.
- design on demand – We create and validate your HPC software solution, including validation on data.
- integration – We fit the application into your software chain.
Publications
- W. Verachtert et al. "Multi party privacy preserving machine learning for process control", Proceedings Volume 11611, Metrology, Inspection, and Process Control for Semiconductor Manufacturing, (2021)
- C. Herzeel et al. "Multithreaded variant calling in elPrep 5", PloS ONE, (2021)
- N. Kurbatova et al. "Urinary metabolic phenotyping for Alzheimer's disease", S. Scientific Reports, (2020)
- C. Herzeel et al. “elPrep 4: A multithreaded framework for sequence analysis”, PloS ONE, (2019)
- P. Costanza et al. A comparison of three programming languages for a full-fledged next-generation sequencing tool, BMC Bioinformatics, (2019)
- E. Smets et al. Large-scale wearable data reveal digital phenotypes for daily-life stress detection. Smets, npj Digital Medicine, (2018)
- D. Decap et al. Parallel variant calling from transcriptomic data using MapReduce. Plos One, (2017)
