By John Russell
May 28, 2009 | The roster of speakers at last week’s Network Biology 2.0 symposium, sponsored by Gene Network Sciences (GNS) and the Broad Institute and held at the Broad, featured some of the most influential people in systems biology and personalized medicine – Jim Collins, George Church, Mark Boguski, and Peter Sorger to name a few. Eric Schadt was scheduled but a last minute conflict caused a switch and he was ably stood in for by his former Merck/Rosetta colleague Jun Zhu.
Clearly it’s good to have friends in high places as GNS leveraged its distinguished SAB and collaborators. The symposium covered a wide swath of topics but in broad terms most talks looked at one or another approach for taking diverse data, transforming that into models, validating those models, and using the model to infer new insight. While solid science dominated the day, there was a more philosophical panel discussion after dinner in which the search for predictive biological models as a practical reality was challenged (more below).
GNS co-founder and executive vice president Iya Khalil chaired the conference. The detailed and highly mathematical nature of most of the talks makes summarizing them here somewhat cumbersome. GNS plans post the slides which would be well worth scanning for researchers with network biology interest. A complete list of the talks is at the end of the article.
The full lineup of speakers included Paul McDonagh, VP of Discovery Biology, GNS; Sorger, professor of systems biology, Harvard Medical School; Gustavo Stolovitzky, manager, functional genomics and systems biology, IBM; Collins, professor of biomedical engineering and & co-director of the Center for BioDynamics, Boston University; Gurinder “Mickey” Atwal, member, Institute for Advanced Study; Dana Pe’er, assistant professor of biology & computer Science, Columbia University; Zhu, Rosetta Inpharmatics/Merck; Boguski, Harvard Medical School & Department of Pathology, Beth Israel Deaconess Medical Center; and Church, Harvard Medical School.
Here are a few of the highlights.
• Examining extrinsic ligand-induced death pathways, Sorger used modeling at different levels – quantitative (ODE) and Boolean – to distinguish which of three potential hypothesis explain why some cells dies rapidly and other slowly. The modeling, perhaps somewhat unexpectedly, ruled out genetics & epigenetics and stochastic biochemical reaction rates. Sorger offered a third hypothesis, suggesting unequal concentrations of the reactants in different cells were the drivers of variability. There was a suggestion from the audience that variability in the number of mitochondria per cell might contribute since the apoptotic pathway examined involved binding to and disrupting mitochondrial membranes.
• Use of Bayesian approaches was a common theme, Pe’re of Columbia showed two pieces of work inferring changes in gene regulatory networks which drove phenotype. In one case, she applied Conexic, a Bayesian algorithm, to a melanoma dataset comprising 62 tumor samples and correctly identified most known ‘driver’ events, while also connecting these to their known targets. The analysis also suggested a number of novel drivers, including a number of genes involved in regulation of protein trafficking and endosome biology in this malignancy. She said preliminary experimental validation supports several of the findings.
• The GNS platform, REFS (reverse engineering/forward simulation) is touted for being able to take large data sets and without prior knowledge use computational techniques to find novel associations. McDonagh presented work in which REFS was used to analyze liver gene expression, serum lipid profiles and body weight measured on 120 male mice from a mouse intercross population. An ensemble of 1024 networks gave accurate predictions of animals that were not part of the training data and explained almost twice the variance compared to quantitative trait loci alone. Further in silico experiments identified an additional 38 transcripts predicted to play a significant role in controlling HDL and free triglycerides plasma concentrations.
• Stolovitzky is co-organizer of the DREAM (dialogue on reverse engineering and assessment methods) program in which participants try to infer networks and targets from blinded data sets. He reviewed a few lessons learned from the challenge sets to date. (For background on DREAM, click http://www.bio-itworld.com/pb/2009/04/16/dream3-update.html) Interestingly, he said, participants who relied heavily on “their favored protein binding sites” did poorly; those incorporating data from the literature did better; and it was possible to “combine” results of the top performers to obtain even better results. Contact Stolovitzky for more details about the upcoming DREAM 4 challenge and how you might participate at email@example.com.
Stepping outside the “network biology-mostly” box, Boguski and Church delivered wide-ranging, optimistic talks on biomedical progress and personalized medicine.
In his presentation, Customized Care 2020, Boguski argued precision diagnostics will be a widespread disruptive innovation in medicine by 2020, and that clinical pathology will play a central role this new era of personalized medicine. Market pressures and advancing science and technology will prompt the changes. “Pathology is moving beyond diagnostics and classification of disease to providing customization information and therapeutic recommendations.”
This is but a small piece and it’s worth watching his webcast video interview with Bio-IT World editor-in-chief, Kevin Davis (click: http://www.bio-itworld.com/LSW/Mark_Boguski). Data-driven reverse engineering of disease processes will identify the underlying molecular pathways that are most relevant in individual patients and these pathways will then be forward-simulated in the presence of virtual drug combinations to predict which therapies will be most effective for individual patients, he argued.
Church delivered the late afternoon keynote with a sweeping vision scarcely contained in its title, “Integrating Personal Genomes, Stem Cell Epigenomes, and Omic Responses to Environments.”
The abstract is perhaps the best summary: “We stand on the cusp of a remarkable opportunity for connecting data for personalized medicine via web 2.0 volunteerism, personally controlled health records, and inexpensive personal genomes. Next-generation sequencing (e.g. http://Polonator.org -- open-source hardware, software, wetware) has brought costs down by over 10-fold per year for 4 years (from $100M to $5K), and with another 100-fold in progress and improvements in interpretability coming, including analysis of regulatory variants via allele specific RNA quantitation by sequencing, and environmental components via pharmaceutical, microbiome and VDJ-ome datasets. Direct whole chromosome haplotypes help establish causative links and improve association studies. http://PersonalGenomes.org is a unique effort to integrate personal genomes with comprehensive sets of medical and non-medical traits and environmental measures and share these for network analyses in an open-access format.”
One hopeful Church suggestion of particular interest to the audience was that rapid advances in transforming skin cells to pluripotent cells will soon expand access to many tissues and enable the development of tissue-specific molecular profiles (gene expression, proteomics, etc).
One of the most interesting and unexpected moments of the day occurred at the symposium’s concluding panel - The Future of Integrative Genomics and Network Biology – with Atwal, McDonagh, Stolovitzky, and Boguski on hand. After a day full solid science and optimism, the second speaker on the panel, Atwal expressed resolute skepticism that systems biology could ever fulfill its promise of predictiveness. You could have heard a pin drop in the conference room as spirits that had been inflated by a glass of wine or beer over a light dinner were suddenly sobered by this challenging view.
A physicist by training, he argued strongly that the search for biology’s first principles around which to organize data and create truly predictive models was probably fundamentally flawed. He recited several long-term research projects which have made, essentially, no progress. He hoped he was wrong, but fundamentally, he didn’t think the field would fulfill its expectations.
McDonagh took on the role of defender, and Boguski, whose 2020 talk brimmed with rosy forecasts for medicine, treaded somewhat lightly in rallying counter argument. Stolovitzky offered that progress would be measured as the iterative process of experiment, model, and simulate gradually moved our understanding closer to comprehensive principles. There was a good deal of audience participation, and no starry-eyed defenders of imminent success.
Usefully, perhaps, the full assembly seemed to recognize the road to comprehensively predictive models for biology is unlikely to be short, even against the backdrop of solid science presented optimistically throughout the day and aiming for that very goal. But if Atwal saw the predictive biology glass forever half-empty, most saw it half-full and rising, though with varying ideas of how full it could get and when.
All in all, it was a very interesting day - let’s hope GNS and Broad do it again next year.
** List of talks: Collins, “Bacterial network biology”; Atwal, “Population Genetics of the Human MDM4 Oncogene: Evolution and Cancer Risk”; Pe’er, “Driving Mutations: Lessons from Yeast and Cancer”; Zhu, “Integrating diverse data to elucidate multi-level regulations of biological systems: A systems approach for complex human diseases”; Boguski: “Customized Care 2020”; Church, “Integrating personal genomes, stem cell epigenomes, and omic responses to environments”
This article first appeared in Bio-IT World’s Predictive Biomedicine newsletter. Click here for a free subscription.