Dr. Irizarry's Opus

When he was a child growing up in Puerto Rico and taking his first tentative plinks at the family piano, Rafael Irizarry could not have known that his budding love for music would one day steer him toward a career as a biostatistician in public health.

By the time he entered college, Irizarry had mastered a whole batch of instruments — the guitar, piano, and cuatro, as well as the conga and bongo drums. And even though he majored in math, he never forsook his first love: He was still playing rhythm for Latin street bands around San Francisco when he entered Berkeley's PhD program in statistics.

So it was natural, when it came time to pick a dissertation subject and his PhD advisor asked him about his interests, that Irizarry quickly answered "music." His advisor suggested he report to the university's Center for New Music and Audio Technology (CNMAT).

Irizarry, an assistant professor of Biostatistics at the School, remembers thinking at the time, "This is not really smart, doing statistics on music. Is this a professional dead end?" But he went ahead in spite of his misgivings.

Researchers at the CNMAT gave him some recorded music and asked him to try separating the music's harmonic sounds from the nonharmonic. The harmonic parts of a musical note consist of faint overtones layered at precise intervals one on top of another above the actual note. This mix of overtones is one part of timbre, the characteristic sound of a particular musical instrument.

But timbre is also formed by the nonharmonic parts of sounds: a calloused finger scraping over a guitar's steel-wound strings, the rasp of a bow against a cello string, saliva moving through the innards of a horn. "These nonharmonic sounds have no pitch or tone," explains Irizarry. "They're what in statistics we call noise, but that's a pejorative term. They are actually crucial to an instrument's sound. If you take this 'noise' out of a guitar recording, for instance, what's left won't sound like a guitar because it won't have the pluck, the nonharmonic part."

At CNMAT, Irizarry used statistical methods to turn musical sounds into zeros and ones and then, using a computer, learned how to siphon off the nonharmonic "noise" as he broke down the music into its constituent parts (a process called decomposing).

The residuals he was collecting were much in demand by young composers at CNMAT looking for new sounds to work with. But besides piquing the interest of avant-garde composers, Irizarry's sorties into musical decomposition steered him into a field of statistics known as time series, which in turn led him to public health.

Unlike the haphazard samples of observations that most statisticians analyze, time series are non-random, consecutive measurements collected at equally spaced time intervals. Long used in astronomy and physics, time series are now of importance to the biomedical sciences because new technologies are allowing researchers to measure over time such things as fetal heart rates, brain waves, circadian rhythms, and the interactions between vast numbers of genes.

But in order for these immense quantities of data to be useful, someone has to figure out how to manage and analyze them. To Irizarry, these cascades of data are like miles-long melodies. "Lots of things in nature are close to being periodic," he notes. "From what I've learned from music, I can use similar techniques to work with measurements of brain signals, say, or pollution levels over time." He pauses, then adds, "I am discovering music everywhere at the School of Public Health."