BioPharma, Diagnostics

Precision medicine 2.0

“Single mutations or single genomic alterations are only really getting us part of the way to a solution,” said John Quackenbush, director of the Center for Cancer Computational Biology and a keynote speaker at MedCity CONVERGE, on the state of precision medicine.

Abstract 3D network in future

Data may be king when it comes to precision medicine, but it’s nothing without its pragmatic queen: biology.

That’s the takeaway from a chat with John Quackenbush, a professor at the Dana-Farber Cancer Institute and director of its Center for Cancer Computational Biology.

In a phone interview ahead of his keynote presentation at the upcoming MedCity CONVERGE conference in Philadelphia, July 31- August 1, Quackenbush discussed the over-simplification of precision medicine and what’s required for science to progress to the next level.

“A lot of the precision medicine we’ve been doing in oncology has really been focused on identifying individual mutations or individual variations of the genome, like HER-2 amplification, and matching patients based on the presence or absence of those single alterations to a therapy,” Quackenbush explained.

That’s a testament to our scientific understanding of the molecular underpinnings of cancer. Yet it also highlights how little we truly know. Despite our grasp of this science, many targeted therapies only work a fraction of the time. And even the most successful drugs of this new era, such as checkpoint inhibitors, don’t universally work as expected.

“What we’ve done in a lot of ways in implementing precision medicine, is we’ve taken a very simplistic view of how biological systems function,” Quackenbush said.

presented by

Attend MedCity CONVERGE to hear from healthcare innovators like John Quackenbush, a professor at the Dana-Farber Cancer Institute and director of its Center for Cancer Computational Biology. Use promo code MCNPOST to save $50. Register now.


It’s a lesson oncology is coming to grips with now. And in time, other fields will follow.

“I think they’re all going to come to the same point in really recognizing that single mutations or single genomic alterations are only really getting us part of the way to a solution,” he said.

So where will the other parts come from?

A lot of recent buzz has centered around the integration of phenotypic data into genomic databases. In other words, tracking the physical characteristics of patients and their tumors over time, not simply making projections from their genomes. The greatest source of this information is medical records, which have recently been digitized in the U.S.

Quackenbush does see a role for this data, but he’s cautious about reading too much into it.

“I think it all has to go together,” he said. “It would be great to pull more data out of EHRs. The problem with most of them is that they’re not designed for research. They’re not even really designed for patient care; they’re designed for reimbursement.”

It’s a challenge to extract information at a macro scale, without losing the context of the data and a firm grasp of its limitations. 

A lot of people are rushing forward to try to interpret this sort of data in an agnostic way, he said. That’s when scientists run into the problem of over-fitting or under-fitting their models. Or they build models that find correlations that lack real world value or relevance.

“I’m somebody who, throughout my career, has always embraced these large-scale datasets,” he noted. “But I’ve always been cautious about how we use them.”

Quackenbush points to a site called “Spurious Correlations” that parodies the leaps data scientists can make when they’re not thinking critically.

For example, author Tyler Vigen found a strong correlation between marriages in Kentucky and national fishing boat deaths, using data from the CDC and National Vital Statistics Reports. Another chart highlights a link between the number of films Nicholas Cage has appeared in and the number of people that drowned by falling into a pool.

It’s safe to say these links are probably not causal. It’s harder to make that call when dealing with complex biology — and the stakes are much higher. To gain real biological insights, Quackenbush said the complexity of human biology needs to be front of mind as huge volumes of data are crunched.

“I see great promise in the future, but only if we’re bold about how we tackle these biological systems.”

Photo: from2015, Getty Images