Biomarkers of Cerebral Cavernous Angioma with Symptomatic Hemorrhage (CASH) - Supplemental

点击翻译按钮获取中文摘要

SUMMARY Cerebral cavernous angioma (CA) is a capillary microangiopathy affecting more than a million Americans, predisposing them to a premature risk of brain hemorrhage. Fewer than 200,000 cases who have suffered a recent symptomatic hemorrhage (CASH) are most likely to re-bleed again with serious clinical sequelae, and are the primary focus of trial readiness and therapeutic development. Candidate biomarkers are emerging from ongoing mechanistic and differential transcriptome studies, which enhance the diagnosis and prediction of CASH, influence clinical decisions, and help stratify high-risk cases in clinical trials. An ongoing project (R01NS114552) has assembled leading clinical CA researchers to deploy advanced computational biology approaches, including supervised machine learning (ML), to discover and validate novel candidate biomarkers. It aims to determine the best clustering and weighing of combined biomarker contributions for optimal diagnostic and predictive accuracy. Initially aimed at combining levels of candidate proteins and microRNAs, recent discoveries have compelled the inclusion of circulating metabolites with mechanistic links, which demonstrate strong diagnostic and prognostic associations in discovery cohorts. The best weighted combination of plasma molecules will be tested in a large validation cohort already recruited, to assess their relevance in sex, age, relevant clinical subgroups and multiple recruitment sites. The project tests a novel integrational approach of biomarker development in a mechanistically defined cerebrovascular disease with a relevant context of use. While applicable to other neurological diseases, the implementation of our data for ML analyses, and its use and usability by other investigators, are limited by inconsistent structure of shared data in current repositories. Furthermore, there is a lack of harmonization of data elements or intuitive connectivity of multiomic data elements from the same patients. This problem is amplified with the addition of metabolites to our multiomic biomarker candidates. In response to NOSI NOT-OD-23-082, we have assembled a team with expertise in computational biology, clinical biomarker research, data science and statistics, who will work on solutions to address these issues. First, we will share each type of raw data in structured repositories that match each type of assay and data type. We also designed a database under best practices for data structure, standardization, and naming conventions that we will share through Dryad, which will include data that is ready for ML and other AI and Deep learning implementations. We further propose creating a GitHub repository that will include a detailed description of the data in Dryad and the codes used for the ML implementation in our study. We will connect the different submissions to the repositories to facilitate the use of more than one data type from the same subjects. Lastly, in addition to publishing the primary results of ML multiomic data analyses from this project, we propose to publish a methods paper on connected shared data structure for readiness in future ML, artificial intelligence and deep learning analyses by other investigators.