HPC and AI at Argonne Help Researchers Chase Down COVID-19 Virus Variants
The COVID-19 pandemic may be over, but many people are still watching nervously as the virus continues to spread and mutate. Can artificial intelligence help us deal with this ongoing global health problem? The answer is yes, and a team of researchers won a Gordon Bell Prize for their work using AI to track virus variants.
They used the Polaris supercomputer at Argonne National Laboratory’s ALCF for their research, along with NVIDIA’s GPU-accelerated Selene system and the Cerebras AI hardware accelerator. ALCF systems, including Polaris, are equipped with workload orchestration by Altair® PBS Professional®, which will also handle the load for Argonne’s Aurora exascale system when it comes online later in the year.
Misbehaving Mutations and Machine Learning
It’s difficult to predict how mutations will behave – whether they’ll evolve to be deadlier or more easily spread from person to person, looking out for their own interests rather than ours. What we can do to fight back and save lives is to act as quickly as possible to track what are known as ‘variants of concern’.
This makes for a very large, complex computational workload. The research team used AI, machine learning, and a year’s worth of genome data for the project.
‘A key challenge in this problem is dealing with long sequence lengths and tackling these foundation models at the scale of the viral genome’. —Argonne computational biologist Arvind Ramanathan
Machine learning played an important role in the research. The team trained large learning models (LLMs) and analysed 1.5 million complete, high-quality SARS-CoV-2 genome sequences, which they used for modelling and to develop the first genome-scale language model (GenSLM) — streamlining a process that would have previously taken much longer and been far more labour-intensive.