On October 25, 2019, the National Heart, Lung, and Blood Institute (NHLBI) convened “Harnessing the New Frontier of Imaging Genomics Workshop for Heart, Lung, Blood and Sleep (HLBS) Disorders,” in Bethesda, Maryland.
This workshop brought together a multi-disciplinary group of scientists with expertise in imaging, genomics, and informatics/data sciences, in addition to biologists and clinicians researching HLBS disorders to address the following objectives:
The workshop is responsive to NHLBI Strategic Vision Objectives 1, 2, 3, 4, 5, and 7.
Imaging Genomics is a new frontier that integrates novel informatics/data sciences approaches, including specific patterns of gene expression and regulation and other omics, on a system-wide scale, with non-invasive radiographic imaging and methodologies that elaborate unique features of endophenotypes (or morphology) and physiology. Imaging Genomics has shown promise in deciphering how genetic variations are associated with imaging endophenotypes to further our understanding of the molecular basis of HLBS disorders. This workshop brought together a multi-disciplinary group of scientists with expertise in imaging, genomics, and informatics/data sciences, in addition to biologists and clinicians researching HLBS disorders. The workshop provided a forum to present the latest advances, cutting-edge approaches, and methodologies in this newly evolving field of Imaging Genomics.
The area of imaging genomics presents many important opportunities for advancement because of the generation of millions of structural and functional images of human tissues, organ systems and the whole body that are available worldwide. Frist, samples with both images and whole genome sequencing are now available and large enough to discover and investigate effects of specific genes on diseases. For example, the NHLBI Trans-Omics for Precision Medicine (TOPMed) program has sequenced genomes from over 150,000 individuals and generated over 877 million genomic variants, along with other -omics data from individuals. Second, innovative computational methods are emerging that extract features in an image for deep phenotyping, facilitating the study of effects of genetic variations. Third, medical technology has advanced to a degree that imaging traits are now measurable with rigor and reproducibility. Finally, such imaging traits are highly heritable and can be acquired in very large numbers of individuals.
Rapid progress has been made using Imaging Genomics for discovery and disease screening, diagnosis, prognostication, prediction, and clinical decision making in a wide range of diseases, such as cancer and neurological diseases. With the rapid development and technical advances in this field, researchers are starting to apply such integrative approaches to HLBS disorders. Several recent studies have shown the feasibility and promise for using Imaging Genomics in HLBS disorders, particularly the use of echocardiography or magnetic resonance imaging (MRI) to track disease traits to identify genetic risk factors in cardiovascular diseases (Arnett et al., BMC Med Genet, 2009; Vasan et al., JAMA, 2009; Fox et al., Circ Cardiovasc Genet, 2013; Marvao et al., J Cardiovasc Magn Reson, 2016; Wild et al., J Clin Invest, 2017).
This workshop demonstrated that Imaging Genomics is an inter-disciplinary field. The speakers demonstrated high proficiency and expertise in three major fields, namely imaging, genetics and artificial intelligence-data science (AI). A major barrier in the Imaging Genomics field is researchers’ need for multi-disciplinary engagement, integration of knowledge across the three domains, and novel tools and resources. It is relatively difficult to have a single individual with high proficiency in all three fields. However, it is possible that programs that enhance collaborations and cross-training programs may help forge these important collaborations and serve to advance Imaging Genomics.
Participants expressed the view that there is a need for better phenotyping, and imaging can play a critical role. Imaging can reveal important endophenotypes, providing additional data that other approaches to phenotyping cannot provide. For example, lung imaging has generated new endophenotypes for chronic obstructive pulmonary disease (COPD). Participants identified an opportunity to collect and build large imaging datasets in databases that are compliant with Findable, Accessible, Interoperable, and Reusable (FAIR) principles.
Robust data-driven approaches could enhance the scientific value of Imaging Genomics research. These approaches benefit from the combination of genomic and imaging data from large populations to facilitate research on the genetic causes of phenotypes. For instance, if an identified genetic variant were associated with endothelial (blood vessel wall) function, then a possible next step would be to use imaging methods to assess vascular permeability. Such an imaging method could be developed but currently is not in routine use in large-scale research projects.
Participants suggested that Imaging Genomics research could focus on imaging in a high-risk group or in a group with strong genetic signatures of risk. Populations that are at high risk and with deeper phenotyping should be a priority to study, as such research may have a better chance of yielding important new knowledge. Among the high-risk groups, molecular imaging could play an important role, though this approach should be strongly hypothesis-driven, given the cost associated with molecular imaging. Multiplexed human molecular imaging was identified as an opportunity in this field.
Future opportunities include combining metabolomics and proteomics data, in addition to genomics data, with state-of-the-art imaging traits to improve precision phenotyping. Participants also identified research opportunities related to longitudinal measures of proteomics/genomics/metabolomics and environmental exposures, along with repeated measures of imaging phenotypes.
Some participants cautioned about the challenge of algorithm overfitting, meaning that the algorithm developed only fits a specific dataset and is not generally applicable to other datasets. This challenge points toward opportunities to develop methods and algorithms to avoid overfitting.
Participants expressed the view that imaging can be a very powerful approach to phenotyping, but there is still a challenge in understanding the potential causal relationship between the genes and the phenotypes. Multi-omics data linked with imaging data could contribute to our understanding of disease and disease processes. Additional opportunities include integration of multi-omics data with imaging data, deep learning, multi-cohort integration, whole genome sequencing and whole exome sequencing , and application of cloud-based platforms or distributed computation.
A white paper outlining the state of the art in this field and highlighting research opportunities that arose from the deliberations of the workshop is in preparation.
NIH Organizing committee: