Simultaneous dimension reduction and adjustment for confounding variation
Dimension reduction methods, such as Multidimensional Scaling (MDS) and Principal Component Analysis (PCA), are commonly applied in high-throughput biological datasets to visualize data in a low dimensional space, identify dominant patterns and extract relevant features. Confounding factors, technically or biologically originated, are commonly observed in high throughput biological experiments.
In this work, we extend Principal Component Analysis to propose AC-PCA for simultaneous dimension reduction and Adjustment for Confounding variation. We demonstrate the performance of AC-PCA and its advantage over existing methods through its application to a human brain development exon array dataset, a model organism ENCODE (modENCODE) RNA-Seq dataset, and simulated data. In the human brain dataset, we found that principal component-based visualization of the neocortical regions is affected by confounding factors, likely originating from the variations across individual donors. As a result, a) it is challenging to identify patterns of the neocortical regions, and b) it is challenging to identify genes associated with interregional variation. In contrast, applying AC-PCA to the human brain dataset, we are able to recover the anatomical structure of neocortical regions and reveal temporal dynamics that existing methods are unable to capture. In the modENCODE RNA-Seq dataset, the variation across different species makes the identification of conserved developmental mechanisms challenging.
Our proposed method is able to capture the shared variation among species and identify genes with consistent temporal patterns in D. melanogaster (fly) and C. elegans (worm) embryonic development. Our proposed method can be applied to more general settings, as demonstrated with simulated examples. We also extended AC-PCA with sparsity constraints for variable/gene selection and better interpretation of the PCs.
Publication
In this work, we extend Principal Component Analysis to propose AC-PCA for simultaneous dimension reduction and Adjustment for Confounding variation. We demonstrate the performance of AC-PCA and its advantage over existing methods through its application to a human brain development exon array dataset, a model organism ENCODE (modENCODE) RNA-Seq dataset, and simulated data. In the human brain dataset, we found that principal component-based visualization of the neocortical regions is affected by confounding factors, likely originating from the variations across individual donors. As a result, a) it is challenging to identify patterns of the neocortical regions, and b) it is challenging to identify genes associated with interregional variation. In contrast, applying AC-PCA to the human brain dataset, we are able to recover the anatomical structure of neocortical regions and reveal temporal dynamics that existing methods are unable to capture. In the modENCODE RNA-Seq dataset, the variation across different species makes the identification of conserved developmental mechanisms challenging.
Our proposed method is able to capture the shared variation among species and identify genes with consistent temporal patterns in D. melanogaster (fly) and C. elegans (worm) embryonic development. Our proposed method can be applied to more general settings, as demonstrated with simulated examples. We also extended AC-PCA with sparsity constraints for variable/gene selection and better interpretation of the PCs.
Publication
- Lin Z, Yang C, Zhu Y, Duchi JC, Fu Y, Wang Y, Jiang B, Zamanighomi M, Xu X, Li M, Sestan N, Zhao H†, and Wong WH†: Simultaneous dimension reduction and adjustment for confounding variation. Proceedings of the National Academy of Sciences of the United States of America. 2016, 113 (51): 14662-14667. [paper link] [software link]