Activity Number:
|
657
|
Type:
|
Contributed
|
Date/Time:
|
Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
|
Sponsor:
|
Section on Statistics in Genomics and Genetics
|
Abstract #320801
|
View Presentation
|
Title:
|
Quadratically Regularized Functional Canonical Correlation Analysis and Its Application to Genetic Pleiotropic Analysis of Multiple Phenotypes
|
Author(s):
|
Nan Lin* and Yun Zhu and Fen Peng and Jinying Zhao and Momiao Xiong
|
Companies:
|
and Tulane University and The University of Texas Health Science Center at Houston and Tulane University and The University of Texas Health Science Center at Houston
|
Keywords:
|
QFCCA ;
CCA ;
Deep learning ;
genetic pleiotropic ;
big data ;
functional data analysis
|
Abstract:
|
Next-generation sequencing and modern biosensing technologies produce dozens of millions of SNPs and large number of phenotypes. A key issue for learning the intricate genotype-phenotype structures is how to effectively extract a few informative internal representation and features from extremely high dimensional genotype and phenotype data. Deep learning uses statistical tools to extract information and reduce the size of "Big Data". Both genome and phenotype signals are compositional hierarchies, in which low level features are combined into higher-level features. Deep hierarchical representations of the genome and phenotype data transform the internal representation at one level into a representation at a higher and more abstract level, extract better informative features and capture invariant features that are only sensitive to the phenotype variability. Motivated by deep learning, we develop a novel quadratically regularized functional canonical analysis (QFCCA) for genetic pleiotropic analysis and apply it to UK-10K dataset. The results show that the QFCCA substantially outperforms all existing statistics for genetic pleiotropic analysis.
|
Authors who are presenting talks have a * after their name.