Validating the performance and inclusivity of a novel functionally-informed predictive genetic test method for polygenic disease

项目摘要 我们现在知道，超过90%的常见疾病的致病变异，包括心血管疾病和 许多癌症位于基因组的非编码区。因此，一个完整的图片常见疾病的风险 需要分析整个基因组的变异，而不仅仅是编码基因组。目前可用的方法 筛查常见疾病风险集中在单基因编码变体上，并且在准确性，包容性， 通过不将功能性全基因组变体纳入其训练方法和输出， 解释。因此，仍然存在未满足的使用系统性全基因组功能作图的需求， 建立风险预测模型。借鉴他们在哈佛、斯坦福大学的研究和经验， 麻省理工学院，布罗德研究所，西奈山，马丁格尔实验室公司。团队开发了第一个全面的 多基因疾病风险模型。这种新的多基因预测模型结合了 新的全基因组功能变异注释与贝叶斯监督机器学习（ML） 预测方法，以提高准确性，种族包容性和预测性基因检测的可解释性。的 这个第一阶段项目的目标是准备Martingale Labs的全基因组预测机器学习 通过验证和量化模型的预测准确性和种族包容性， 与目前临床上可用的方法相比。这些模型可以在临床环境中实施 作为预测性基因测试，以帮助根据风险对个体进行分层，并制定预防策略， 和预防性药物，以尽量减少疾病的风险。在这个项目中，我们关注的是心血管疾病的例子， 疾病我们将在第二阶段将模型性能的量化扩展到其他常见疾病， 从乳腺癌前列腺癌和结肠直肠癌开始以目前的测试速度，即使是保守的 通过目前的健康保险公司的报销，我们提出的基因检测产品可以获得80亿美元的利润。 年度收入机会。我们的项目涉及开发一种新型的深层技术， 第一个包含全基因组功能注释的监督学习模型。这 技术用途从医学应用扩展到其他新颖和有用的应用，例如动物 模特或兽医

PROJECT SUMMARY We now know that over 90% of causal variants for common diseases, including cardiovascular disease and many cancers, lie in non-coding regions of the genome. Therefore, a complete picture of common disease risk requires analyzing variants across the whole genome, not just the coding genome. Currently available methods to screen for common disease risk focus on monogenic coding variants and are limited in accuracy, inclusivity, and interpretability by not incorporating functional whole-genome variants into their training methods and output interpretation. Thus, there is still an unmet need for the use of systematic whole-genome functional mapping in building predictive risk models. Drawing on prior research and experience from their time at Harvard, Stanford, MIT, the Broad Institute, and Mount Sinai, the Martingale Labs, Inc. team has developed the first comprehensive functionally informed polygenic model for disease risk. This new polygenic prediction model combines the power of novel genome-wide functional variant annotations with a Bayesian supervised machine learning (ML) prediction method to improve the accuracy, ethnic inclusivity, and interpretability of predictive genetic tests. The goal of this Phase I project is to ready Martingale Labs’ whole-genome predictive machine learning platform for scale by validating and quantifying the model’s predictive accuracy and ethnic inclusivity as compared to current clinically available methods. These models can be implemented in clinical settings as predictive genetic tests to help stratify individuals by risk and tailor preventative strategies such as screenings and preventative medications to minimize disease risk. In this project, we focus on the example of cardiovascular disease. We will expand the quantification of our model performance to other common diseases in Phase II, starting with breast, prostate, and colorectal cancer. At the current rate of testing, even with conservative reimbursement by current health insurers, our proposed genetic testing product could capture an $8 Billion annual revenue opportunity. Our project involves the development of a new type of deep technology by creating the first supervised learning models that incorporate functional annotations of the whole genome. This technology’s use extends beyond medical applications to other novel and useful applications, such as animal models or veterinary medicine.