Workshop in Biostatistics
MSOB, Room 303
|DATE:||March 8, 2018|
|TITLE:||Methods for interpreting genetic variation of unknown significance|
||Nilah Monnier Ioannidis, Postdoctoral Research Fellow, Biomedical Data Sciences
Research Engineer, Biomedical Data Science-Administration, Stanford
Understanding the clinical significance of personal genome variation is a major challenge for personalized medicine, with large numbers of variants of unknown significance discovered in next-generation sequencing studies. I will first discuss two machine learning tools that we recently developed to predict the clinical significance of individual genetic variants. REVEL is a random forest that predicts the pathogenicity of missense variants, with an emphasis on rare missense variants found in whole exome and whole genome sequencing studies. FIRE is a set of random forests that predicts whether single nucleotide variants, including both coding and noncoding variants, are likely to regulate the expression levels of nearby genes. In addition to individual variants, I will also discuss the complementary approach of analyzing combinations of genetic variants to predict their collective effects on complex traits and intermediary phenotypes. By predicting personal gene expression levels from personal genetic variation using the recently proposed prediXcan linear regression models, we conducted a transcriptome-wide association study of cutaneous squamous cell carcinoma, a common form of skin cancer, to identify genes whose predicted expression levels are associated with disease risk. I will discuss advantages and disadvantages of this approach as well as our ongoing work on developing improved methods for predicting the effects of personal genome variation on gene expression patterns and linking them to downstream effects on diseases and other complex traits.
Tools for single variant interpretation:
1. Ioannidis NM, Rothstein JH, et al. REVEL: An ensemble score for predicting the pathogenicity of rare nonsynonymous variants. American Journal of Human Genetics 99(4):877-885 (2016).
2. Ioannidis NM, et al. FIRE: Functional inference of genetic variants that regulate gene expression. Bioinformatics 33(24):3895-3901 (2017).
Background on prediXcan and transcriptome-wide association studies (TWAS):
1. Gamazon ER, et al. A gene-based association method for mapping traits using reference transcriptome data. Nature Genetics 47(9):1091-1098 (2015).
2. Wainberg M, et al. Vulnerabilities of transcriptome-wide association studies. bioRxiv 206961 (2017).