It’s human nature to forget history — your own as much as the history that ends up in books. I was reminded of a little Forrest Gump moment in my own history recently when I came across this article about Johns Hopkins and their work on creating an in-house coronavirus test.
Hospitals are the front line for dealing with outbreaks. What most people don’t realise is that because they are on the front-lines, hospitals are often called upon to create their own tests to identify a pathogen long before a kit-based test from a vendor comes to market.
Many years ago, not long after we started to see anthrax posted around the country, HSARPA (Homeland Security Advanced Research Projects Agency) had recognised that pathogens could be either an intentional or unintentional threat to national security and had started looking at technologies that could help us quickly identify pathogens. The challenge was, that in order to identify a pathogen, you had to isolate that pathogen, which often took several days. During which time, the pathogen could continue to be spread. What was needed was a means of identifying a pathogen without having to isolate it from its host.
A friend of mine asked me to help out on a project he was working on at the University of Houston. He had invented a technology that allowed you to identify the genomic signature of a pathogen regardless of the host. These host-blind signatures could be used by hospitals to create tests that could be run in any path lab using off-the shelf technology. You can read more about it here.
The challenge was one of scale. The original algorithm that was designed to work on a single Windows computer, and would run for hours as it read through genomes, looking for sequences that were unique to the pathogen. You could always throw beefier hardware at the problem, but invariably you would hit the memory and performance limits of the machine. You needed a way to be able to quickly generate these signatures, and create a database of signatures that agencies around the US could use to create their own in-house tests.
The first challenge was to take the very spare algorithm, written in Microsoft C++, and make it run on the university’s Linux cluster. This involved a bit of refactoring, some documentation, a bit of unit testing, learning C++ and a lot of cursing. But ultimately, with some help from the folks at the High Performance Computing lab, we were successful.
If we were to tackle the same challenge today, the approach would probably be very different. The cloud computing technologies and GPUs that are now ubiquitous, would make short work of analysing large genomic data sets, not to mention memory-mapped IO and NoSQL databases.
You don’t often get a chance to work on code that could literally save someone’s life, but this was one of those rare opportunities, and one that I’ll be eternally grateful for.