Isaac Kohane - US grants

Pathology specimens and their associated clinical data, archived in repositories of various configurations, represent a vast and underutilized mine of valuable resource. The emergence of the Internet and its related technology now-offers an opportunity to coordinate these valuable resources. Cost, logistical, and public policy issues make a centralized repository of specimens and/or specimen-related information unpalatable. In "Consented High-performance Indexing and Retrieval of Pathology Specimens" (CHIRPs) we propose an approach to a distributed specimen-informatics network, that will allow institutions to maintain local control of specimens, related identifiers and other sensitive information while safely sharing anonymized data across institutions. CHIRPs will support a novel peer-to-peer network where each site will announce their presence to others, and will distribute queries among themselves, in a manner similar to GnutellaNet. One key addition to the model is a method for secure authentication of clients and servers when needed, through the use of digital certificates. These major goals will be addressed: 1) Establishing a scalable Extensible Markup Language (XML) representation of specimen annotation that will support both a least common denominator access and an advanced query access to existing specimen information across multiple healthcare delivery and research institutions, 2) Formulating a taxonomy of patient consent that is part of the XML-based annotation, allowing for a balance between the advancement of biomedical knowledge and the protection of patient privacy, and 3) Developing a peer-to-peer distributed architecture for indexing and searching for specimens that leverages the Internet and the Web, and that minimizes the effort needed to participate in the Shared Pathology Informatics Network (SPIN). CHIRPs will be implemented and tested for its scalability across the Harvard/UCLA consortium, which joins two large academic medical centers, each with an established comprehensive cancer center, the Dana- Farber/Harvard Cancer Center and the Jonsson Comprehensive Cancer Center. The Harvard/UCLA consortium is composed of 9 member institutions representing 7 different pathology information systems and their associated archives containing millions of annotated specimens. The development and implementation of caws will provide a means to harness these existing valuable resources and a generalized platform to index and host specimen-related information prospectively to support future collaborations.

DESCRIPTION (provided by applicant): With the increased use of genetic testing, whether through broadening of testing criteria or the well documented ordering of tests by physicians because of patient demand, and furthermore the commoditization of measurements of thousands of variants on a single individual, the number of healthy individuals undergoing genetic testing is accelerating. The growth in the number of false positives is projected to rise dramatically because many of the current annotations for known mutations have not been developed for the asymptomatic, well population. With the false positive results, patients will be alarmed unnecessarily, and also clinicians will order unnecessary and occasionally risky tests. Furthermore, insurance companies will find the growth in unnecessary secondary testing triggered by this tsunami of false positive tests ("the incidentalome") as an unanticipated financial risk and thereby endanger the very real benefits that the sound use of genetic testing and long-anticipated genomically-enable "personalized medicine" can bring. We propose to demonstrate that this risk of the incidentalome is substantial by automatically mining the biomedical literature, by scanning public genomic data sets, and computationally predicting the effects of mutations to identify a set of candidate mutations that are predicted to not contribute to disease in the general population. This despite their annotation in authoritative genetic databases and texts as "highly penetrant" in causing disease congenitally or in childhood. Two thousand adult patients known not to have these diseases will be then tested for each of these candidate mutations with the hypothesis that we will demonstrate the presence of some of these mutations in these health individuals. Evidence supporting this hypothesis will both serve as an important caution in the use of genetic testing and will also demonstrate the value of population studies to find the broader clinical significance of genetic mutations.

DESCRIPTION (provided by applicant): We propose to create a massively scalable toolkit to enable large, multi-center Patient-centered Information Commons (PIC) at local, regional and, national scale, where the focus is the alignment of all available biomedical data per individual. Such a Commons is a prerequisite for conducting the large-N, Big Data, longitudinal studies essential for understanding causation in the Precision Medicine (1) framework while simultaneously addressing key complexities of Patient Centric Outcome Research studies required under ACA (Affordable Care Act). Our proposal is solidly grounded in our experience over the last 25 in harnessing clinical care data to the research enterprise. In creating PIC we propose to focus on: 1. Enable the identification and retrieval of all data that pertain to individal health by creating a data sharing architecture that is capacious enough for all relevant data types and that enables patient and institutional autonomy to be respected. 2. Test fully-scaled implementations of the proposed architecture early in the development process, with the active involvement of a committed user community that seeks to use allowed us to refine our designs to facilitate subsequent robust dissemination and adoption. 3. Provide commodity workflows that can be used to 'clean' and complete the often noisy and sparse data gathered in the course of observational studies. 4. Embrace decentralization while enabling the construction of a nationally or regionally-scaled patient-centered information commons. 5. Encourage the selection of standards through the tools that enable the construction of patient-centered information commons. 6. Employ diagnostic classification and prognostication as figures of merit to measure how well a patient-centered information commons adds the understanding of patient populations. In addition to the research and development agenda we have also taken on the development of educational opportunities for end user community to become more familiar with the methods and challenges of data science.

The Administrative Component of the Patient-centered Information Commons (PIC) will be responsible for managing the overall conduct of this project. The management plan, headed by an Executive Team that includes the co-PIs, Drs. Isaac Kohane and Shawn Murphy, and the Executive Director, Susanne Churchill, has proposed a model based on this group's ten year experience with the i2b2 U54 National Center for Biomedical Computing. This plan focuses on the development and sustenance of a truly interactive and collaborative working group that involves all of the interdisciplinary domains required by this big data science project. Regularly structured interactions, progress reporting and ongoing evaluation are proposed to insure that progress is monitored, challenges identified, and solutions devised to address bottlenecks. Leadership will rely on a number of advisory bodies, including the NIH Science Team, an internal Scientific Advisory Board with expertise in areas affecting but not directly proposed for our research (e.g., patient privacy), an External Advisory Committee to be configured after award, and very importantly, a Users' Group constituted from potential end users in the community. The Admin Team will be responsible for developing and maintaining a dissemination strategy for the open source tools and procedures emerging from our work. Significant effort will be devoted to assuring compliance with all financial, regulatory and reporting requirements. This team will be fully engaged in the DB2K Consortium activities, including participation in its advisory bodies, dedicated meetings and other activities still to be defined.

DESCRIPTION (provided by applicant): As a result of the accelerated pace of development of technologies for characterizing the human genome, the rate-limiting step for large scale genomic investigation in clinical populations is now phenotyping. This is particularly the case for neuropsychiatric (NP) illness, where phenotypes are complex, biomarkers are lacking, and the primary cell types of interest are difficult to access directly. It has become apparent that both rare and common genetic variation contributes to disease risk and that this risk crosses traditional diagnostic boundaries in psychiatry. Taking advantage of a large, already-established NP biobank could dramatically accelerate progress toward understanding the cross-disorder mechanism of action of disease liability genes. This study proposes novel applications of emerging technologies in informatics and cellular neurobiology to eliminate this phenotyping bottleneck. In doing so, it will accelerate investigation of clinical and cellular phenotypes for understanding single and multilocus/polygenic associations. Aim 1: Adapt and expand one of the largest NP cellular biobanks by parsing electronic health records with gold-standard assessment of cognition and other RDoC phenotypes. Aim 2: Define the genome-wide multidimensional functional genomics (MFG) landscape in NP disease into which the transcriptomic signature (RNA-seq) of each induced neuron (IN) representing a clinically characterized individual is projected. The projection provides the mapping from molecular to phenotypic characterization and a directionality towards healthful/neurotypical states used in Aim 3. Aim 3: Develop a probabilistic model of gene expression dependencies that will predict which small molecular perturbations are likely to shift the IN transcriptomic signature in a healthful direction in the MFG and to then update the model based on measured perturbations in the MFG. Aim 4: Select patient samples to study in greater detail for epigenetic (DNA methylation, histone marks and RNA editing) and transcriptional control particularly with regard to activity dependent changes that have been implicated in many NP diseases. Aim 5: Here we assess just how well the clinical phenotypes are informed by the genome-wide characterizations and assess which is more robust.

The Training Component of the Patient-centered Information Commons or PIC has chosen to focus on three major elements that will rely (1) on the strength of this team's existing infrastructure at the Center for Biomedical Informatics at Harvard Medical School and (2) the new science proposed for the Data Science Research component of this proposal to support the overall goals of the Big Data to Knowledge initiative. Direct training of the next generation of leaders is offered in two forms, a pre-doctoral-level distributed training initiative and an undergraduate research internship. With the goal of attracting students to the field of big data science, the competitive Distributed Pre-doctoral Program will target students currently enrolled in quantitatively-focused graduate programs across the country who have passed their qualifying exams and would like to engage in a distance collaborative project with faculty at PIC, thereby exposing them to opportunities not available at their local schools. The undergraduate research internship (Summer Institute in Bioinformatics and Integrative Genomics) will offer a nine week, intensive immersion in didactic lectures with leading big data scientists and a mentored research project with PIC faculty. A second major element will develop a series of instructional Big Data videos that will be publically available to the community. Choice of topics will be developed in consultation with the Consortium members. Lastly, the PIC training and science teams will host both an annual Big Data Conference and a series of monthly Lectures which will be available to the community via videography (for the Conference) and WebEx (for the Lecture series). Success of these initiatives will be evaluated by a defined set of metrics, including surveys and outcomes assessment.

Although, the Center Consortium Component of the BD2K initiative will not be fully delineated until after all awards are made, the Patient-centered Information Commons (PIC) Center plans to be fully engaged in all elements of this collective should we be selected to join. We have already been in conversation with several other applicants and understand that there are several proposals with overlapping and synergistic goals that would be highly desirable to collaborate on. Meaningful focus on toolkit development in the area of ontology loading/translation; GPU-accelerated algorithms, particularly those used in the imputation and linkage techniques described in our Aims 2 and 3; convenience functions relating to regulatory compliance and confidentiality; and data visualization would all be very synergistic with our proposal and provide a valuable resource to complement our efforts. We are pleased to be able to offer back to this community early access to testing of our toolkit, educational opportunities and products, and a cloud hosted PIC that will contain a large data set of synthetic patients for use with local and national PICs (this is made possible by an in kind contribution from industry).

DESCRIPTION (provided by applicant): As a result of the accelerated pace of development of technologies for characterizing the human genome, the rate-limiting step for large scale genomic investigation in clinical populations is now phenotyping. This is particularly the case for neuropsychiatric (NP) illness, where phenotypes are complex, biomarkers are lacking, and the primary cell types of interest are difficult to access directly. It has become apparent that both rare and common genetic variation contributes to disease risk and that this risk crosses traditional diagnostic boundaries in psychiatry. Taking advantage of a large, already-established NP biobank could dramatically accelerate progress toward understanding the cross-disorder mechanism of action of disease liability genes. This study proposes novel applications of emerging technologies in informatics and cellular neurobiology to eliminate this phenotyping bottleneck. In doing so, it will accelerate investigation of clinical and cellular phenotypes for understanding single and multilocus/polygenic associations. Aim 1: Adapt and expand one of the largest NP cellular biobanks by parsing electronic health records with gold-standard assessment of cognition and other RDoC phenotypes. Aim 2: Define the genome-wide multidimensional functional genomics (MFG) landscape in NP disease into which the transcriptomic signature (RNA-seq) of each induced neuron (IN) representing a clinically characterized individual is projected. The projection provides the mapping from molecular to phenotypic characterization and a directionality towards healthful/neurotypical states used in Aim 3. Aim 3: Develop a probabilistic model of gene expression dependencies that will predict which small molecular perturbations are likely to shift the IN transcriptomic signature in a healthful direction in the MFG and to then update the model based on measured perturbations in the MFG. Aim 4: Select patient samples to study in greater detail for epigenetic (DNA methylation, histone marks and RNA editing) and transcriptional control particularly with regard to activity dependent changes that have been implicated in many NP diseases. Aim 5: Here we assess just how well the clinical phenotypes are informed by the genome-wide characterizations and assess which is more robust.

The Management and Training Core represents the arm of our Center of Excellence in Genomic Science Neuropsychiatric Genome-Scale and RDOC Individualized Domains (N-GRID) proposal that will be responsible for insuring that the specific aims proposed in the scientific proposal, including the educationaland training objectives, are realized in the most effective manner possible. The administrative structure will be led by the Project PI, Dr. Zal< Kohane in collaboration with the Executive Director and Dr. Roy Perlis, lead on the psychiatric and neuronal cell line aspects of the research. Together these individuals constitute the Executive Team and will be responsible for insuring that an interactive and collaborative working team environment is built and sustained and that maximum flexibility to build on results and availability of new techniques/technologies is insured. The proposed management model is based on a nine year experience by this team on a comparably complex, multidisciplinary, high risk U54 and will include the following elements: required full team weekly working group meetings to insure coordination of all goals and appropriate fonward progress on each; dedicated fixed day a week for CEGS business, including Executive Team Meeting to review current status, identify any outstanding roadblocks and determine allocation or reallocation of resources as necessary; a strong Executive Director who will closely monitor all aspects of the program to insure maximum real time flexibility; and an External Advisory Committee that will meet at least annually and whose input will be regularly solicited. This Core will also be responsibly for developing a training program for our postdoctoral fellows that is tailored to each person's background and interests. The N-GRID CEGS will participate in the existing Harvard Medical School Diversity Action Plan (DAP) by actively recruiting new postdocs and summer students, providing an expanded roster of cutting edge genomicists to mentor these trainees, and othenwise insure the professional advancement of minorities underrepresented in the genomic sciences.

DESCRIPTION (provided by applicant): As William Harvey wrote, Nature is nowhere accustomed more openly to display her secret mysteries than in cases where she shows traces of her working apart from the beaten path; nor is there any better way to advance the proper practice of medicine than to give our minds to the discovery of the usual laws of nature by careful investigation of cases of rarer forms of diseases. In addition to the missed scientific opportunity, marginalization of those with undiagnosed conditions exacts a tremendous human toll on health, emotional well-being, family life, and finances. As the Coordinating Center for a new Undiagnosed Diseases Network, Harvard Medical School will support the expansion of the National Institutes of Health's Intramural Undiagnosed Diseases Program into a collaborative, integrated network of 5-7 additional clinical sites with the goals of developing effective protocol for the diagnosis and care of people with undiagnosed conditions, as well as advancing the study of undiagnosed conditions by sharing high quality laboratory and clinical data generated by the network. We will do this by leveraging our collective expertise and experience in managing multi-site studies, trans-institutional data sharing, data analysis, biomedical curation, diagnosing and caring for those with rare conditions, and performance review to serve as the scientific, organizational, and operational foundations of the integrated and collaborative research community across the network sites. In so doing, we will accomplish five aims: (1) Provide the infrastructure, support, curation, and scientific leadership for the principled harmonization of clinical protocols, data standards, data analysis, and quality improvement practices throughout the network; (2) Create infrastructure to identify and match patients and clinical teams with intra- and extra-network expertise and resources; (3) Establish content and infrastructure to support transitions in care following UDP evaluation for diagnosed and undiagnosed patients; (4) Perform and support network dissemination activities; and (5) Continually evaluate the performance of individual clinical sites as well as the network as a whole.

We propose to create a massively scalable toolkit to enable large, multi-center Patient-centered Information Commons (PIC) at local, regional and, national scale, where the focus is the alignment of all available biomedical data per individual. Such a Commons is a prerequisite for conducting the large-N, Big Data, longitudinal studies essential for understanding causation in the Precision Medicine framework while simultaneously addressing key complexities of Patient Centric Outcome Research studies required under ACA (Affordable Care Act). This agenda entails the four following aims: Aim 1: Create an individual patient data identification and retrieval toolkit that is robust across distributed data of wide variety and geographically scattered. Robustness with regard to a variety of organizational structures and national scalability is emphasized. Aim 2: Generate a complete diagnostic and prognostic 'data' picture of a patient across multiple sources of data, some of which are noisy and sparse. Aim 3. Enable robust decentralized computation on large-scale data with the Patient-centered Information Commons Big Data Science Platform (PIC-DSP), particularly in configurations where data are generated in locations other than where computational resources are most available. Aim 4: Create three patient-centered information commons instances (PICIs) to test all aspects of the toolkit developed. We have selected neurodevelopmental disorders as our first PICI, as it fulfills several criteria (wide variety of data types and scales, collaborator engagement, multiple healthcare institutions, and opportunity to rigorously test and refine features of the tool).

DESCRIPTION (provided by applicant): It is intuitive that the genetic risk for human disease depends on the environment, or that the effect of an exposure in disease is not identical across human populations of different genetic backgrounds. This concept is known as gene-by-environment interaction (GxE) and it is hypothesized that disease risk can be better explained by identifying GxE. Despite the importance in understanding GxE in human disease, there have been few studies that have documented the concept. There are a number of explanations for few-recorded GxE. First, there a few ways to measure standardized indicators of the environment (unlike single nucleotide polymorphisms [SNPs]). When GxE are investigated, environmental factors are selected without sufficient evidence of their prior association in disease traits. Second, investigating GxE requires large sample sizes to identify interactions between individual SNPs and environmental factors. The problem is exacerbated when accounting for multiple tests of millions of SNPs with small main effects. Using current day methods and unstandardized environmental data, it is difficult to collect evidence for interactions between millions of specific SNPs and environmental factors. It is now possible to detect GxE in complex disease traits that contribute to significant disease burden, such as body mass index (BMI) and blood pressure (BP), by developing new methods in quantitative genetics and leveraging existing methods in environmental exposure bioinformatics. This project has four aims to achieve this goal. First, the investigators will develop and validate genome-wide polygenic prediction scores to summarize the contribution of all common SNPs in BMI and BP. The investigators will develop and validate the scores in preexisting genome-wide association study (GWAS) consortia data. In the second aim, the investigators will standardize environmental variables from 7 independent cohort studies deposited in the Database of Genotypes and Phenotypes (dbGaP) to build a large cohort of N ~ 30K for GxE testing. Third, the investigators will develop methods to detect and validate GxE between polygenic trait scores and specific environmental factors selected from Environment-Wide Association Studies (EWAS) in BMI and BP with the combined dbGaP cohorts. Fourth, the investigators will estimate the proportion of variation in BMI and BP due to GxE interaction. The methods proposed in the R21 provide a new paradigm for GxE estimation by taking advantage of all SNPs on the genome while considering a larger number of environmental factors with robust support from EWAS. This will lead to a more complete picture of variability ascribed to genes and environment in complex traits of highest disease burden. If successful, the methods will enable the rapid documentation of reproducible GxE, a need in the human genetics and environmental health fields.

PROJECT SUMMARY In pursuit of a diagnosis, many patients undergo countless tests and procedures in hopes of finding answers. When these fail to yield diagnoses, patients are left in a state of uncertainty, faced with the possibility of never knowing the cause of their symptoms. Finding a diagnosis and other patients affected with the same, or similar, condition can reduce or eliminate this uncertainty and end the diagnostic odyssey. In our current healthcare system, it can take years before patients with rare conditions and rare presentations of common conditions receive a diagnosis- an amount of time that many of these patients simply do not have. As the Coordinating Center for the Undiagnosed Diseases Network (UDN), Harvard Medical School will support the integrated network of 8-10 clinical sites and multiple core laboratories striving to improve the level of diagnosis and care for patients with undiagnosed conditions and facilitate research into their etiologies. We will do this by leveraging our collective expertise and experience in managing multi-site studies, trans-institutional data sharing, data curation and analysis, bioinformatics, biostatistics, and translational research. In so doing, we will accomplish five aims: (1) Enhance UDN core operational processes, (2) Develop consensus across the UDN for evaluation and diagnostic approaches, (3) Follow the UDN participant beyond the diagnostic evaluation, (4) Expand the transactional and analytical intelligence applied to UDN data, and (5) Integrate a larger set of scientific and clinical care partners into the UDN.