Bits to Bedside: Using Big Data to Create Better Precision Medicine

Tom Cahoon:

Hi, everyone. Welcome to this podcast from Cambridge Health Institute for the 2017 Molecular Medicine and Tri-Con, which runs from February 19 to the 24 in San Francisco, California. I'm Thomas Cahoon, Market Research Analyst at Cambridge Innovation Institute. We have with us today one our speakers from the bioinformatics track, Doctor Dexter Hadley, Assistant Professor of Pediatrics at the Institute of Computational Health Sciences at the University of California, San Francisco. Doctor Hadley, thank you for joining us.

Dexter Hadley:

Thank you for having me.

Tom Cahoon:

Your first talk, you'll be discussing how you use large data sets to create better personalized care. What kinds of diseases have you studied with this data and if it's an app, how has your mobile app helped bring better medicine to patients?

Dexter Hadley:

I have been lucky, or fortunate enough, to be working with large data sets for quite some time now. Since I was in of medical school that has been the goal to computationally use data to help affect better medicine. I started out for my PhD thesis when I was finishing medical school at Penn. I used genetic data to look at autism and neuropsychiatric disease that resulted in a clinical trial for ADHD and autism and a precision medicine clinical trial. I mean, we've identified a small proportion of patients that we think will benefit with very high response rate to a specific drug, so a personalized approach to autism, ADHD. Since then I've expanded my use of big data. We did some work on Dengue Fever. Dengue is the most significant viral disease of our time, kills 20,000 people every year throughout the world. We think we can use big data to help decrease that rate of fatality.

I've not only looked at private data from patients, but also this idea of open data that funding agencies have been supporting for the last 15, 20 years. We find that to be extremely useful. You asked me about web applications. I have a web application and an application to take public data and engage a large crowd of people that may be interested in using that data. It may not have the bioinformatics or the technical abilities to really leverage the data, but can ask very insightful, biological questions that we have classes on using this resource called star geode, a search tag analyze resource for the gene expression omnibus. We motivate people who would not necessarily be using this data to help plan their experiments, oppose hypotheses that they can go validate in the lab, really an effort to democratize the data.

I have a web application to do that. Finally, more recently, I've got in digital health as much of the world is adopting smart phones and Apple watches and so on and so forth. We have an application to use smart phones to screen for melanoma, for instance, using sophisticated computer vision algorithm as a technique. I don't discriminate. I think data is frozen knowledge and applying data, either generated in private clinical care settings, like a hospital, generated in research settings, things put on the internet as open data, or even prospectively with either on the internet or mobile, is a very prolific way to come up with new ways to manage and think about and treat disease.

Tom Cahoon:

Your second talk is about developing and learning health care systems for detecting melanoma. How do you see this progressing and benefiting cancer diagnosis, treatment, and research in the future.

Dexter Hadley:

Right. The statistics are something like a person dies from melanoma every hour throughout the world. Yet, if we catch melanoma early enough, 97% of people survive. There's a huge discordancy there, meaning that we do a pretty poor job. There's little precision in making a diagnosis. Is this a benign mole or is this skin cancer, i.e. melanoma? I see computers as providing an objective way to make such a diagnosis, not only objectively, but actually cell phones is a convenient way to make such a diagnosis given the power of computers to make an objective diagnosis. Right now, the way dermatologists characterize melanoma is purely subjective. We all learn about various acronyms.

The ABCDE method is probably the most famous one where you look at the asymmetry of a mole, the border of a mole, the color, how it's evolved over time, and so on and so forth. Very subjective measures that, for instance, is not like a glucose measurement is for diabetes. That's a very objective measure. How big the mole is or how irregular the border of the mole is is very subjective. I see computers bringing a level of objectivity and precision into making the diagnosis to begin with. We've all heard recently artificial intelligence is coming to the fore now. Google is training its chief learning algorithms to beat Starcraft, for instance. It's recently beat the world champion for Go, a Chinese equivalent of chess. The computers are outperforming humans regularly these days, and I'd like to see that kind of super human performance applied to medicine.

The key in what drives that artificial intelligence, self-driving cars, whatever amazing artificial intelligence you have, guarantee there's a lot of data behind it to support the algorithms and to support that type of fluid decision making. Now our idea is very simple. We generate a lot of data in the clinics, period. More so for medical, legal reasons and billing reasons, and anything else. It would be good to translate that data that is being generated in enormous quantities into actual patient care and management. Every biopsy for melanoma is photographed for medical reasons, for billing reasons, for instance. It's coded. We are just taking that a step further and training computer vision algorithms to not replace, but to augment the objectivity in making these diagnoses more convenient, make them sooner, and hopefully change the rate at which people are dying from skin cancer.

Tom Cahoon:

Lastly, what are you hoping to learn at this conference?

Dexter Hadley:

This is an interesting conference. It's in San Francisco, the Molecular Medicine Tri-Conference. I think it's great to get together a lot of diverse people that are needed to evoke this change and what I consider a data-driven change. I work in the agency for Computational Health Sciences at UCSF. It really is all about the data for me. I think putting people like me together with people that actually make clinical decisions - I don't actually see patients. I have been through medical school. I have gone through residency, but I don't actually see patients. Putting people together with me that have the skills to evoke this change with people that have the skills to implement the change in the clinic I think is a really powerful opportunity.

It's nice to know what clinicians are looking for. It's nice to learn about the problems that clinicians have, sort of the inprecision they face, the tools, the tools that they would like, and how they see this data revolution that I think is ongoing, how they see this playing out in their careers. I think it's great to have conferences like these that bring a lot of different and diverse minds and thinking together to try to build some kind of synergy, to really take advantage of a revolution in data that we're generating at breakneck speed today in clinical medicine.

Tom Cahoon:

Thank you for joining us today.

Dexter Hadley:

Okay, thank you for having me.

Tom Cahoon:

That was Doctor Dexter Hadley, Assistant Professor of Pediatrics at the Institute for Computational Health Sciences at the University of California, San Francisco, speaking about his bioinformatics talk from Bits to Bedside at the upcoming 2017 Molecular Medicine Tri Con, taking place February 19 through 24 in San Francisco, California. If you're interested in this event and would like to learn more, go to triconference.com for registration information. I'm Thomas Cahoon. Thank you for listening.


Premier Sponsors

NeoGenomics

Seegene