Multivariate visual clustering of single nucleotide polymorphisms and clinical predictors using Chernoff faces

Abstract

With advanced technology, collection of health-related data in undertaken on a large scale, producing large and high-dimensional data. Visualization of such data is important and useful for further statistical analyses such as classification and clustering. However, visualizing large multivariate datasets is challenging, especially for high dimensional data, as they are often complex and confounded. Currently, visualization for Single Nucleotide Polymorphisms (SNPs) and clinical predictors of disease are assessed separately. As there is increasing evidence of genetic-environmental interactions for pregnancy complications, prediction models based solely on either clinical measurements or genetic risk factors may be inadequate. Hence, we present an example of multivariate visualization on combinations of clinical measurements and SNPs through Chernoff faces, and perform visual clustering for prediction of Preterm births (PTB). A random sample containing 100 patients (Uncomplicated pregnancy = 92, PTB=8) with 11 clinical and 4 genetic predictors are visualized into faces with various style of eyes, ears, nose and hair, showing two groups with similar face characteristics amongst Uncomplicated pregnancies and Preterm births. The faces identified as PTB appear to have either a tall hair style or no ears, which correspond to whether the mother was herself born preterm, and SNPs in TGFβ and IL1β genes.