By John Russell
June 27, 2008 | Human hearing is wonderfully discriminating. Most of us effortlessly notice when a musical instrument drifts out of tune or a piece of music breaks out of harmony. The ‘bad’ notes stick out. What’s more, the simple tones we hear are actually many overtones blended together – in this sense music acts as a kind of natural integrator of complex data.
Gil Alterovitz, a research fellow at Harvard/MIT Division of Health Science and Technology (in Children's Hospital Informatics Program) and a research affiliate, MIT Computer Science and Artificial Intelligence Laboratory, has been pioneering efforts to turn data – gene expression levels, for example – into musical applications that could be used in biomedical research, diagnostics, and for an endless variety of monitoring systems. To a significant extent the secret sauce is made of clever algorithms for data compression and transduction into notes.
You might, for example, construct a four-note chord, in which each of the notes is built from very many data points, weighted in various ways so that as data values increase or decrease, individual notes could sharpen, flatten, or become completely dissonant. The changing pitches, abrupt or gradual, could tell a lot about the state of underlying process being described, and in near real-time if data is captured quickly enough. Surgeons in an operating room could easily follow some set of parameters as they vary, while still performing surgery.
“If one thing goes off, that doesn’t mean that the alarm really should go off; and what happens many times now [in operating rooms] is many people are ignoring the alarms,” says Alterovitz. “With this method a number of things have to go wrong for you to really start noticing that there is a problem. And if a number of things are going wrong it gradually will sound inharmonious and capture this fuzzy nature of it.”
Sonic signatures can be useful in biomedical research, and Alterovitz has done so in a colon cancer study.
“It is impossible to keep track of all of the genes at any given time,” he says. “Each gene is one dimension. So, 10 thousand genes have 10 thousand dimensions. It is also possible to represent a group of genes in fewer dimensions by using mathematical compression. In the colon cancer case, I condensed 3,142 genes to four linear combinations.”
Here is Alterovitz’ explanation: Each linear combination is a different dimension of those of 3,142 genes. Each linear combination is assigned a different note.
Note one = collection of all 3,142 genes in a particular dimension 1.
Note two = collection of all 3,142 genes in a particular dimension 2.
Note three = collection of all 3,142 genes in a particular dimension 3.
And so on, for a total of four different notes (representing four different dimensions of the same 3,142 genes).
“So, when playing three or more notes at the same time, we get chords. When played across time, we get music,” explains Alterovitz. “Using Pythagorean tuning mathematics, we assigned the normal to be harmonious. So, then, deviations (like disease) turn out to naturally sound inharmonious. Thus, when we listen to harmonious music, it means the person is healthy; when we listen to inharmonious music it could be a sign of a pathological problem.”
In the colon cancer case, the growing inharmonious sound represented changes in gene expression (up or down) over time. In a diagnostic or treatment setting, training your ear might catch cancerous changes more quickly than other pathology methods. Alterovitz envisions many diverse applications for this sonification of data
“When someone goes into shock, there are all of these cytokines that get released in the blood stream. We look at some of these end products now, but if you could look at them much earlier [it would be helpful], or when someone has a stroke there are all these particles released that actually cause more damage than the actual stroke after a certain amount of time. If we could see ahead of time, because genes are kind of like the blueprint, their expression determines ahead of time what will happen. It will allow us to monitor earlier what is going on, and that way we can have a treatment plan before [the problem] actually occurs,” says Alterovitz.
Next Step: Commercialization
If you smell a new company emerging, you’re right. Alterovitz and his colleagues have been talking to venture capitalists in recent months about how best to commercialize the technology. Children’s Hospital is also looking at some applications. One question is how to handle the intellectual property rights.
“That was an interesting discussion. We have this possibility of applying for a patent within one year of a certain date. But when we were talking to the venture capitalists, they were saying that we may not want to do a patent, we may want to keep it as a trade secret, the actual code and algorithm details, the fine details of it, because once you apply for a patent, everything is out there and anyone can just write a program. Epecially if you use something very efficient like MATLAB, you could write it very quickly.”
Indeed, Alterovitz had praise for the widely used MATLAB from The MathWorks. He stressed that its ability to link easily to many databases “is very useful, particular when you need to collaborate with experts in music, who did not necessarily have expertise in database querying and management.”
So far there’s been no shortage of ideas for potential applications. Cockpits in airplanes, for example, might benefit from a “musical monitoring” system. He was contacted by someone wondering if this might be applied in submarine sonar systems, making them easier to interpret.
One point of discussion has been how to generate cash flow from the start – a seeming prerequisite for getting funding nowadays.
“At the beginning you don’t really have a product, necessarily, right? But there are some applications that may not be as medically inspiring, but will bring in a revenue stream,” says Alterovitz. “For example, I’ve had many people express an interest in getting music, related to either their DNA information or even if it’s not their own genetic information, just general gene-based music. There’s a DJ who we’re working with who wants to play it in the Boston dance clubs around here. There are companies that license this type of material. That is something that can kind of quickly bring in cash flow and also potentially some branding for [us].”
Rock on!