Jun 19 2017

It’s Benign!

by at 2:09 pm

My surgeon Dr. Shawna Willey walked into the patient exam room where I waited nervously. I first saw her thumbs up before her beaming face. I could breathe again!

My friends and I who recently turned 40 and starting our baseline mammograms can’t help but wonder about the lack of consensus on optimal cancer screening strategies, target populations, its benefits and harms. My colleague Dr. Jeanne Mandleblatt and her team have studied breast screening strategies for decades and have shown that biennial screening from ages 50-74 achieves a median 25.8% breast cancer mortality reduction whereas annual screening from ages 40-74 reduces mortality an additional 12% but introduces very high false positive rates. Many women and their families are subject to extreme anxiety due to the sheer number of repeat mammograms, false-positives, benign biopsies and in 7% of the cases an over diagnosis.

My experience and of my friends with breast cancer screening are raising many questions. How can we better predict the target risk population who must undergo screening early and often? Would this decision-making process consider risk factors, lifestyle, and patient preferences? How often are patients with a diagnosis of a benign breast condition on a stereotactic core needle biopsy upgraded to a non-benign diagnosis on an excisional biopsy which requires full sedation and surgery? What was the care journey like for other patients like me – Asian female, healthy, no family history? How many in the US and globally have access to the excellent care and follow-up that I was privileged to receive from Dr. Willey and her expert team?

Touted as the fourth industrial revolution, Artificial Intelligence is poised to empower clinicians, patients and researchers in answering these questions. What is AI? The term was coined by Dartmouth professor Dr. John McCarthy in 1956 and defined as “the science and engineering of making intelligent machines, especially intelligent computer programs.” Applications of AI in medicine have been limited by the complexity of highly cognitive processes such as making a medical diagnosis or selecting a treatment which require integration of thousands of datasets with millions of variables and multiple interactions between these variables. It takes years to collect, organize and publish practice changing results such as Jeanne’s screening study. What if we could use data that we routinely collect during the care process and effectively use AI to assist clinicians in real-time to make informed treatment decisions?

Interested in learning more about AI in Biomedicine? Want to engage with expert scientists and product developers in AI? Register for Georgetown’s Big Data in Biomedicine symposium on October 27th!

Companies like Google and Amazon are betting big on this. Jeff Bezos wrote “..it is hard to overstate how big of an impact AI will have on society over the next 20 years”; Google’s Sundar Pichai, when asked recently about the next big thing at Google responded “I can’t quite tell exactly but advances in AI and machine learning, we are making a big bet on that and this will bring a difference in many many fields”.

We cannot have a conversation about AI in medicine without discussing IBM Watson, the supercomputer that sifted through 20 million cancer research papers, and conducted a differential diagnosis on a difficult to treat leukemia patient in 10 minutes by combining genomic data with the power of cognitive computing. One concern that informaticians including my informatics mentor Dr. Bill Hersh have raised is that the publicity around Watson has mostly been from news articles and press releases, primarily from researchers at IBM and call for a more scientific analysis, not n-of-one case reports, of its abilities in clinical decision making. Systems like Watson will benefit from systematic expert knowledge input to guide the cognitive computing processes in navigating the complex medical pathways.

While still early, AI is already starting to make important contributions to Medicine says AI professor at MIT and a recent breast cancer survivor, Dr. Regina Barzilay. She and her team are asking all the right questions of data – “can we apply the sophisticated algorithms we use to predict customer’s shoe-buying habits to adjust treatments for cancer patients?” “Can computers detect signs of breast cancer or even pre-malignancy earlier than humans are currently capable of?” And the Holy Grail – “Can we use the huge quantities of data from smart toothbrushes, wearables, genomic sequencing, medical records to get to the first and right treatment?”

What next?

In the last decade, big data in biomedicine has focused on collecting (e.g. through mobile and other IoT) and organizing (e.g. cloud computing) information but all signs point in one direction for the next decade – real world applications of AI. We will witness the development of expert systems, question-answering systems and deep learning methods that begin to address complex real world problems in medicine. These will augment, not replace, human expertise. Winners will find ways to rapidly and accurately integrate human input with computational output. Usability of these tools by end users and human factors will be key.

While a true tech automation enthusiast at heart and practice, I will never forget Dr. Willey’s kind and soft words as she clearly explained my pathology report. She also carefully noted in my medical record the rare chlorohexidine pre-op antiseptic agent hypersensitivity that I had developed post anesthetic induction.

One more data point!

              Let’s continue the conversation:

No responses yet | Categories: From the director's office,Newsletter,Subha Madhavan | Tags: , , , , ,

Jul 07 2016

Bioinformatics is a vocation. Not a job.

by at 10:57 am

Bioinformatics is at the heart of modern day clinical translational research. And while experts define this as an interdisciplinary field that develops and improves methods and tools for storing, retrieving, organizing, and analyzing biological (biomedical) data – it is much, much more!

Bioinformatics helps researchers connect the dots between disparate datasets; improve extraction of signal from noise; predict or explain outcomes; and improves acquisition and interpretation of clinical evidence. Ultimately, it allows us to tell the real data stories.

To effectively tell these stories, and to see this all-encompassing domain in biomedical research and its true super powers, we must pursue bioinformatics as a vocation – or a calling – and not just a job.

Spring’16 has been a busy season for us Bioinformaticians at the Georgetown ICBI. I carefully curated six of our recent impact stories that you may find useful.

  1. AMIA’16 – The perfect triangulation between clinical practitioners, researchers and industry can be seen at AMIA annual conferences. I was honored to chair the Scientific Planning Committee for this year’s AMIA Translational Bioinformatics (TBI) Summits, featuring sessions on the NIH Precision Medicine initiative, BD2K program, and ClinGen. I sat down with GenomeWeb’s Uduak Grace Thomas for a Q&A on this year’s Summit, which attracted over 500 informaticians. Come join us at the AMIA Joint Summits 2017 to discuss the latest developments in Bioinformatics.
  1. Cyberattack Response! – We were in the middle of responding to NIH’s request for de-identified health record data for our Precision Medicine collaborative when MedStar Health, our health care partner’s computer systems, were crippled by a cyberattack virus. Thousands of patient records were inaccessible and the system reverted to paper records, seldom used in modern hospital systems. Thanks to the hard work and dedication of the IT staff, MedStar Health systems were restored within days with no evidence of any compromised data, according to the MedStar Health spokesperson. However, our research team had to act fast and improvise a way to fulfill the NIH’s data request. We ended up providing a complete synthetic linked dataset for over 200 fields. As our collaborator Josh Denny, a leader in the NIH Precision Medicine Initiative put it – “this experience you had to go through will help us be better prepared for research access to EHRs for nationwide clinical networks”. We sure hope so!
  2. Amazon Web Service (AWS) – The AWS Public Sector Summit was buzzing with energy from an active ecosystem of users and developers in federal agencies, small and large businesses, and nonprofit organizations—a community created over just the past few years. It was enlightening for me to participate on a panel discussing Open Data for Genomics: Accelerating Scientific Discovery in the Cloud, with NIH’s Senior Data Science Advisor, Vivien Bonazzi, FDA’s former Chief Health Informatics Officer, Taha Kass-Hout and AWS’s Scientific Computing Lead, Angel Pizarro. Three take homes from the Summit – (1) a growing need for demand-driven open data; (2) concern over the future administration’s commitment (or lack thereof) to #opendata; and (3) moving beyond data storage, and the future of on-demand analytics.
  3. Massive Open Online Course (MOOC) on Big Data – Want to begin demystifying biomedical big data? Start with this MOOC – to be available through Open edX late Fall. Georgetown University was recently awarded a BD2K training grant to develop an online course titled “Demystifying Biomedical Big Data: A User’s Guide”. The course aims to facilitate the understanding, analysis, and interpretation of biomedical big data for basic and clinical scientists, researchers, and librarians who have limited/no significant experience in bioinformatics. My colleagues Yuriy Gusev and Bassem Haddad, who are leading the course, are recording interviews and lectures with experts on practical aspects of use of various genotype and phenotype datasets to help advance Precision Medicine.
  4. Know Your TumorSM – Patients with pancreatic cancer can obtain molecular tumor profiling through the Pancreatic Cancer Action Network’s Know Your TumorSMprecision medicine initiative. It is an innovative partnership with Perthera, a personalized medicine service company that facilitates the multi-omic profiling and generates reports to patients and physicians. Check out the results from over 500 KYT patients presented at AACR’16 by our multi-disciplinary team of patient coordinators, oncologists, molecular diagnostic experts and data scientists.
  5. Moonshot – Latest announcement from VP Biden’s Cancer Moonshot program unveiled a major database initiative at ASCO’16. I had the opportunity to comment in Scientific American on the billions of bits of information that such a database would capture to help drive an individual’s precise cancer treatment. Continue to watch the Moonshot program if you are involved with cancer research or care continuum.

It is personally gratifying to see Bioinformaticians, BioIT professionals, and data scientists continue to solidify their role as an integral part of advancing biomedicine. I have yet to meet a bioinformatician who thinks of her/his work as just a job. Engage your bioinformatics colleagues in your work, we will all be better for it!

One response so far | Categories: From the director's office,Newsletter,Subha Madhavan | Tags: , , , , , , , , , , ,

Feb 13 2015

Informaticians on the “Precision Medicine” Team

by at 8:34 am

My first recollection of the term “Precision Medicine” (PM) is from a talk by Harvard Business School’s Clayton Christensen on disruptive technologies in healthcare and personalized medicine in 2008. He contrasted precision medicine with intuitive medicine, saying, “the advent of molecular diagnostics enables precision medicine by allowing physicians to delineate conditions that are likely constellations of diseases presenting with a handful of symptoms.” The term became mainstay after NRC’s report, “Toward precision medicine: Building a knowledge network for biomedical research and a new taxonomy of disease.” Now, we converge on the NIH’s definition– PM is an emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle.

“Cures for major diseases including cancer are within our reach if only we have the will to work together and find them.  Precision medicine will be the way forward,” says Dr. John Marshall, head of GI Oncology at MedStar Georgetown University Hospital.

The main question in my mind is: How can we apply PM to improve health and lower cost? Many sectors/organizations are buzzing with activity around PM to help answer this question.

NIH is developing focused efforts in cancer to explain drug resistance, genomic heterogeneity of tumors, monitoring outcomes and recurrence and applying that knowledge in the development of more effective approaches to cancer treatment. In a recent NEJM article, Drs. Collins and Varmus describe NIH’s near-term plan for PM in cancer and a longer-term goal to generate knowledge that is broadly applicable to other diseases (e.g., inherited genetic disorders and infectious diseases). These plans include an extensive characterization and integration of health records, behavioral, protein, metabolite, DNA, and RNA data from a longitudinal cohort of 1 million participants. The cost for the longitudinal cohort is roughly $200M to expand trials of genetically tailored treatments, explore cancer biology, and set up a “cancer knowledge network” for sharing this information with researchers and oncologists.

FDA is working with the scientific community to ensure that the public can be confident that genomic testing technology is safe and effective while preserving innovation among developers. The FDA recently issued draft guidance for a framework to regulate laboratory-developed tests (LDTs). Until now, most genomic testing is done through internal custom developed assays or commercially available LDTs. The comment period just ended on Feb 2.

Pharma/Biotech companies are working to discover and develop medicines and vaccines to deliver superior outcomes for their customers (patients) by integrating “Big Data” (clinical, molecular, multi-omics including epigenetics, environmental, and behavioral information).

Providers, health systems, and Academic Medical Centers are incorporating appropriate molecular testing in the care continuum and actively participating in clinical guideline development for PM testing and use.

Public and private Payors are working to appropriately determine clinical utility, value and efficacy of testing to determine reimbursement levels for molecular diagnostic tests – a big impediment for PM testing right now. Payors recognize that collecting outcomes data is key to determining clinical utility and developing appropriate coding and payment schedule.

Diagnostic companies are developing and validating new diagnostics to enable PM, especially capitalizing on the new value-based reimbursement policies for drugs. They are also addressing joint DX/RX approval processes with the FDA.

Professional organizations are setting standards and guidelines for proper use of “omics” tests in a clinical setting – examples include AMA’s CPT codes, ASCO’s QOPI guidelines, or NCCN’s compendium.

Many technology startups are disrupting current models in targeted drug development and individualized patient care to deliver on the promise of PM. mHealth domain is rapidly expanding with innovative mobile sensors and wearable technologies for personal medical data collection and intervention.

As informaticians and data scientists, we have atremendous opportunity to collaborate with these stakeholders to contribute in unique ways to PM:

  1. Develop improved decision support to assist physicians in taking action based on genomic tests.
  2. Develop common data standards for molecular testing and interpretation
  3. Develop methods and systems to protecting patient privacy and prevent genetic discrimination
  4. Develop new technologies for measurement, analysis, and visualization
  5. Gather evidence for clinical utility of PM tests to guide decisions on utility
  6. Develop reference databases on the molecular status in health and disease
  7. Develop new paradigms for clinical trials (N of one trials, basket trials, adaptive designs, other)
  8. Develop methods to bin patients by mutations and pathway activation rather than by tissue site alone.
  9. Create value from Big Data

What are your ideas? What else belongs on this list?

Jessie Tenenbaum, Chair, AMIA Genomics and Translational Bioinformatics shares: “It’s an exciting time for informatics, and translational bioinformatics in particular. New methods and approaches are needed to support precision medicine across the translational spectrum, from the discovery of actionable molecular biomarkers, to the efficient and effective storage and exchange of that information, to user-friendly decision support at the point of care.”

A PricewaterhouseCoopers analysis predicts the total market size of PM to hit between $344B-$452B in 2015. This includes products and services in molecular diagnostics, nutrition and wellness, decision support systems, targeted therapeutics and many others. For our part, at ICBI, we continue to develop tools and systems to accurately capture, process, analyze, and visualize data at patient, study, and population levels within the Georgetown Database of Cancer (G-DOC). “Precision medicine has been a focus at Lombardi for years, as evidenced by our development of the G-DOC, which has now evolved into G-DOC Plus. By creating integrated clinical and molecular databases we aim to incorporate all relevant data that will inform the care of patients,” commented Dr. Lou Weiner, Director, Lombardi Comprehensive Cancer Center who was invited to the White House precision medicine rollout event on January 30.

Other ICBI efforts go beyond our work with Lombardi. With health policy experts at theMcCourt School of Public Policy, we are working to identify barriers to implementation of precision medicine for various stakeholders including providers, LDT developers, and carriers. Through our collaboration with PRSM, the regulatory science program at Georgetown, and the FDA, we are cataloging SNP population frequencies in world populations for various drug targets to determine broad usefulness of new drugs. And through theClinGen effort, we are adding standardized, clinically actionable information to variant databases.

The President’s recent announcements on precision medicine have raised awareness and prompted smart minds to think deeply about how PM will improve health and lower cost. We are one step closer to realizing the vision laid out by Christensen’s talk in 2008. ICBI is ready for what’s next.

Let’s continue the conversation – find me on e-mail at subha.madhavan@georgetown.edu or on twitter at @subhamadhavan

No responses yet | Categories: From the director's office,Newsletter,Subha Madhavan | Tags: , , ,

Sep 13 2013

ICBI Director’s blog post, Fall 2013

by at 4:34 pm

Nate Silver is my new hero. His prediction (well ahead of other political analysts and media outlets) of President Obama’s victory in 2012 exemplifies the various facets of data science – data collection, pre-processing, filtering, analyzing, and presenting information – almost  in real-time.  His simple prediction model and detailed data presentation techniques have inspired and amazed data scientists across multiple domains such as health care, biomedical research, sports analysis, politics, astronomy, and many others. We clearly live in a data-driven economy. If you haven’t gotten enough of the statistics on big data in health care, here are a few more.  U.S. health care data is growing at the rate of 30 petabytes per year. Global health data size is estimated at approximately 150 exabytes, growing at 1.2 to 2.4 EB/year. The potential value of healthcare data, either through pharmaceutical product development dollars or reimbursement gains, is estimated at $300 billion annually.

So, why should we care about all this? As biomedical researchers, we not only curate big data but also play important roles as analysts, interpreters, and decision makers. As costs of big data generation drop, techniques such as targeted and whole-genome sequencing, RNA-Seq, Chip-Seq, miRNA-Seq and others are proving to be quite useful in the identification of novel and rare anomalies associated with disease, gene expression signatures, and functions of non-coding RNAs in tissue and blood. We will take a deeper dive into one of these techniques – RNA-Seq – and review its data analysis challenges and opportunities.

Although RNA-Seq was developed in 2008, the bioinformatics methods to analyze these data continue to evolve. We have come a long way from testing changes in expression of only a few genes using low-throughput techniques such as RT-PCR. The use of microarrays to study gene expression on a genome-wide scale has become the primary high-throughput method to study gene expression over the past decade. Yet this method has many shortcomings, including the inability to identify novel transcripts, a limited dynamic range for detection, and difficulty in reliability and cross-experimental comparisons. RNA sequencing overcomes many of these problems. High-throughput next-generation sequencing methods to sequence the entire transcriptome permit both transcript discovery and robust digital quantitation of gene expression levels.

The bioinformatics tools can be categorized based on the applications of RNA-Seq data and the questions we want to ask of the data. Current applications and related tools are listed below for ease of access. Note that tools and software continue to evolve and improve.

  1. Read mapping – Transcriptome sequencing reads are usually first mapped to the genome or the transcriptome sequences, and read alignment is a basic and crucial step for the mapping-first based analytical methods. The complexities of genome sequences have direct influences on the mapping accuracy of short reads. Large genomes with repetitive and homologous sequences make it difficult to perform short read mapping. Also, as introns and exons vary in length, accurate mapping is necessary to identify true boundaries. Tools for read mapping include among others BowtieBWA, and SOAP2.
  2. Splice junction detection – Alternative splicing is very common in the gene transcriptional process of eukaryotes, and is very important for the genomes to generate various RNAs (both protein-coding and non-protein-coding) to ensure proper molecular functioning. RNA splicing can be described as the primary challenge to correctly map the sequence reads that cover splice junctions to reference sequences. To identify the splice junctions between exons, the software must support spliced mapping for reads, because the reads across the splice junctions need to be split into smaller segments, and then mapped to different exons by crossing-checking with possible introns. Tools for splice junction detection include, among others, TopHatMapSplice, and SpliceMap.
  3. Gene and isoform expression testing – With microarrays, we are limited to quantifying expression only at the gene level. By contrast, RNA-Seq can estimate expression at both gene and isoform level. To comprehensively understand the transcriptome, it is important to study expression at the gene isoform level. RNA-Seq can also help detect unannotated genes and isoforms for any species while microarrays depend on prior information from known genes. Tools for genes and isoform quantitation from RNA-Seq include, among others, Cufflinks, MISO, and Scripture.
  4. Differential expression analysis – RNA-Seq can be used to detect both differentially expressed genes and isoforms, while microarrays are limited for differentially expressed genes. Since genes with multiple exons can encode different functional isoforms, this is an important factor to consider when selecting the proper technologies for research. Although it is still relatively more costly to sequence multiple samples than microarrays, RNA-Seq will inevitably and eventually replace microarrays. While RNA-Seq provides a digital count of genes and isoforms that help quantify expression levels, several RNA-Seq biases should be taken into account such as sequencing depth, count distribution among samples, and length of genes and transcripts. Tools for differential expression analysis from RNA-Seq include, among others, Cufflinks, bayseq, and DESeq.

Once we complete pre-processing and gene expression analysis, a number of downstream analyses can follow depending on the questions we want to answer for that particular dataset. Such analyses may involve functional enrichment, network inference, integration with other data types that will ultimately lead to biological insights, and new hypothesis generation. Software tools such as Ingenuity, Partek, Pathway Studio and many others help with downstream analysis of RNA-Seq data. Tools aggregators or workflow developers such as Globus Genomics combine a number of these tools into readily usable data analysis pipelines.

In addition to basic science applications, RNA-Seq has the potential to become a clinically applicable technology. In disease classification and diagnosis, RNA-Seq could provide a powerful tool for high-resolution genomic analysis of human tissue samples and cell populations to identify novel mutations and transcripts in cancers, to classify tumors based on gene expression patterns, or to identify microbial pathogens based on sequence identification. While the sensitivity of this method lends itself nicely to clinical use, challenges associated with small sample sizes, data analyses and interpretation, and education of clinical personnel must be overcome before it can be broadly used in that setting. Still, the day we will routinely use RNA-Seq and/or similar methods clinically in the practice of precision medicine is not far off. Let’s continue the conversation – find me on e-mail at sm696@georgetown.edu or on twitter at @subhamadhavan.

No responses yet | Categories: From the director's office,Newsletter,Subha Madhavan | Tags: , , , , , ,