Juergen A. Klenk

De-identification of medical images intended for research is a core requirement for data-sharing initiatives, particularly as the demand for data for artificial intelligence (AI) applications grows. The Center for Biomedical Informatics and Information Technology (CBIIT) of the US National Cancer Institute (NCI) convened a virtual workshop with the intent of summarizing the state of the art in de-identification technology and processes and exploring interesting aspects of the subject. This paper summarizes the highlights of the first day of the workshop, the recordings, and presentations of which are publicly available for review. The topics covered included the report of the Medical Image De-Identification Initiative (MIDI) Task Group on best practices and recommendations, tools for conventional approaches to de-identification, international approaches to de-identification, and an industry panel.

Background: As genome sequencing becomes better integrated into scientific research, government policy, and personalized medicine, the primary challenge for researchers is shifting from generating raw data to analyzing these vast datasets. Although much work has been done to reduce compute times using various configurations of traditional CPU computing infrastructures, Graphics Processing Units (GPUs) offer opportunities to accelerate genomic workflows by orders of magnitude. Here we benchmark one GPU-accelerated software suite called NVIDIA Parabricks on Amazon Web Services (AWS), Google Cloud Platform (GCP), and an NVIDIA DGX cluster. We benchmarked six variant calling pipelines, including two germline callers (HaplotypeCaller and DeepVariant) and four somatic callers (Mutect2, Muse, LoFreq, SomaticSniper). Results: We achieved up to 65 × acceleration with germline variant callers, bringing HaplotypeCaller runtimes down from 36 h to 33 min on AWS, 35 min on GCP, and 24 min on the NVIDIA DGX. Somatic callers exhibited more variation between the number of GPUs and computing platforms. On cloud platforms, GPU-accelerated germline callers resulted in cost savings compared with CPU runs, whereas some somatic callers were more expensive than CPU runs because their GPU acceleration was not sufficient to overcome the increased GPU cost. Conclusions: Germline variant callers scaled well with the number of GPUs across platforms, whereas somatic variant callers exhibited more variation in the number of GPUs with the fastest runtimes, suggesting that, at least with the version of Parabricks used here, these workflows are less GPU optimized and require benchmarking on the platform of choice before being deployed at production scales. Our study demonstrates that GPUs can be used to greatly accelerate genomic workflows, thus bringing closer to grasp urgent societal advances in the areas of biosurveillance and personalized medicine.

Background As genome sequencing becomes a more integral part of scientific research, government policy, and personalized medicine, the primary challenge for researchers is shifting from generating raw data to analyzing these vast datasets. Although much work has been done to reduce compute times using various configurations of traditional CPU computing infrastructures, Graphics Processing Units (GPUs) offer the opportunity to accelerate genomic workflows by several orders of magnitude. Here we benchmark one GPU-accelerated software suite called NVIDIA Parabricks on Amazon Web Services (AWS), Google Cloud Platform (GCP), and an NVIDIA DGX cluster. We benchmarked six variant calling pipelines, including two germline callers (HaplotypeCaller and DeepVariant) and four somatic callers (Mutect2, Muse, LoFreq, SomaticSniper). Results For germline callers, we achieved up to 65x acceleration, bringing HaplotypeCaller runtime down from 36 hours to 33 minutes on AWS, 35 minutes on GCP, and 24 minutes on the NVIDIA DGX. Somatic callers exhibited more variation between the number of GPUs and computing platforms. On cloud platforms, GPU-accelerated germline callers resulted in cost savings compared with CPU runs, whereas somatic callers were often more expensive than CPU runs because their GPU acceleration was not sufficient to overcome the increased GPU cost. Conclusions Germline variant callers scaled with the number of GPUs across platforms, whereas somatic variant callers exhibited more variation in the number of GPUs with the fastest runtimes, suggesting that these workflows are less GPU optimized and require benchmarking on the platform of choice before being deployed at production scales. Our study demonstrates that GPUs can be used to greatly accelerate genomic workflows, thus bringing closer to grasp urgent societal advances in the areas of biosurveillance and personalized medicine.

BACKGROUND: Artificial intelligence (AI), including machine learning (ML) and deep learning, has the potential to revolutionize biomedical research. Defined as the ability to “mimic” human intelligence by machines executing trained algorithms, AI methods are deployed for biomarker discovery. OBJECTIVE: We detail the advancements and challenges in the use of AI for biomarker discovery in ovarian and pancreatic cancer. We also provide an overview of associated regulatory and ethical considerations. METHODS: We conducted a literature review using PubMed and Google Scholar to survey the published findings on the use of AI in ovarian cancer, pancreatic cancer, and cancer biomarkers. RESULTS: Most AI models associated with ovarian and pancreatic cancer have yet to be applied in clinical settings, and imaging data in many studies are not publicly available. Low disease prevalence and asymptomatic disease limits data availability required for AI models. The FDA has yet to qualify imaging biomarkers as effective diagnostic tools for these cancers. CONCLUSIONS: Challenges associated with data availability, quality, bias, as well as AI transparency and explainability, will likely persist. Explainable and trustworthy AI efforts will need to continue so that the research community can better understand and construct effective models for biomarker discovery in rare cancers.

As more datasets, tools, workflows, APIs, and other digital resources are produced by the research community, it is becoming increasingly difficult to harmonize and organize these efforts for maximal synergistic integrated utilization. The Findable, Accessible, Interoperable, and Reusable (FAIR) guiding principles have prompted many stakeholders to consider strategies for tackling this challenge by making these digital resources follow common standards and best practices so that they can become more integrated and organized. Faced with the question of how to make digital resources more FAIR, it has become imperative to measure what it means to be FAIR. The diversity of resources, communities, and stakeholders have different goals and use cases and this makes assessment of FAIRness particularly challenging. To begin resolving this challenge, the FAIRshake toolkit was developed to enable the establishment of community-driven FAIR metrics and rubrics paired with manual, semi- and fully-automated FAIR assessment capabilities. The FAIRshake toolkit contains a database that lists registered digital resources, with their associated metrics, rubrics, and assessments. The FAIRshake toolkit also has a browser extension and a bookmarklet that enables viewing and submitting assessments from any website. The FAIR assessment results are visualized as an insignia that can be viewed on the FAIRshake website, or embedded within hosting websites. Using FAIRshake, a variety of bioinformatics tools, datasets listed on dbGaP, APIs registered in SmartAPI, workflows in Dockstore, and other biomedical digital resources were manually and automatically assessed for FAIRness. In each case, the assessments revealed room for improvement, which prompted enhancements that significantly upgraded FAIRness scores of several digital resources.