| Title: | Matrix Analysis and Normalization by Concordant Information Enhancement |
|---|---|
| Description: | High-dimensional data integration is a critical but difficult problem in genomics research because of potential biases from high-throughput experiments. We present MANCIE, a computational method for integrating two genomic data sets with homogenous dimensions from different sources based on a PCA procedure as an approximation to a Bayesian approach. |
| Authors: | Tao Wang, Chongzhi Zang |
| Maintainer: | Tao Wang <[email protected]> |
| License: | GPL-2 |
| Version: | 1.4 |
| Built: | 2024-11-15 03:19:58 UTC |
| Source: | https://github.com/cran/MANCIE |
This function removes noise in the main matrix by utilizing information available from the supplementary matrix or summarized supplementary matrix.
mancie(mat_main,mat_supp,cutoff1=0.5,cutoff2=0)mancie(mat_main,mat_supp,cutoff1=0.5,cutoff2=0)
mat_main |
The main matrix or data frame. Rows are features (genes/peaks/etc) and cols are samples (conditions/replicates) |
mat_supp |
The supplementary matrix or data frame. |
cutoff1 |
The higher cutoff. See below for explanation. |
cutoff2 |
The lower cutoff. See below for explanation. |
If the supplementary dataset have the same genomic features on rows and samples on columns as mat_main, it can be directly fed to mancie. An example is RNA-Seq data of the same cell lines from two labs. If the supplementary dataset has different rows from mat_main. It need to be first summarized using summarize_mat to be compatible with mat_main. An example is RNA-Seq data and DNase-seq data of the same tissue types.
The underlying rationale for using MANCIE is that the variation of genomic features in mat_supp are concordant with and can be used to remove noise in the variation of genomic features in mat_main.
(a) If the correlation between row i of mat_main and row i of mat_supp is larger than cutoff1, the new row vector will be the first PC of the matrix formed by these two row vectors.
(b) If the correlation is between cutoff1 and cutoff2, the new row vector will be the weighted average of these two rows. The weight for row i of mat_main is 1 and the weight for row i of mat_supp is the correlation between these two row vectors.
(c)If the correlation is smaller than cutoff2, the new row vector is the original row i of mat_main
There should be a reasonable portion of rows that fall into the first and second category. If not, the user should check if the data they would like to try MANCIE on really fits the aforementioned rationale. The user may also vary the default values of cutoff1 and cutoff2 if they see fit. The mancie function will report percentage of rows falling into each category.
A modified matrix with the same dimensions as the main matrix
data(mancie_example,package="MANCIE") sum_DNase=summarize_mat(exp,ann_exp,DNase,ann_DNase) lev_exp=mancie(exp,sum_DNase)data(mancie_example,package="MANCIE") sum_DNase=summarize_mat(exp,ann_exp,DNase,ann_DNase) lev_exp=mancie(exp,sum_DNase)
MANCIE package
This demo dataset is a small portion of the Encode dataset used in our publication.
data(mancie_example)data(mancie_example)
4 data frames
Summarize information in the supplementary matrix according to physical location into a new matrix with the same dimensions as the main matrix
summarize_mat(mat_main,ann_main,mat_supp,ann_supp,n_limit=50,extend=100000,method="pca")summarize_mat(mat_main,ann_main,mat_supp,ann_supp,n_limit=50,extend=100000,method="pca")
mat_main |
The main matrix or data frame. Rows are features (genes/peaks/etc) and cols are samples (conditions/replicates) |
ann_main |
|
mat_supp |
The supplementary matrix or data frame. Rows are features (genes/peaks/etc) and cols are samples (conditions/replicates) |
ann_supp |
|
n_limit |
The most number of closet features in the supplemenatry matrix that can be used for summarization for each feature in the main matrix |
extend |
The genomic features in the supplemenatry matrix that are no farther away than |
method |
Which method to summarize the information in the supplementary matrix when there are >1 neighboring row vectors associated with the row vector in the main matrix. "pca" (default) or "max". In the "max" method, the row vector of these neighboring vectors with the highest correlation with the row vector in the main matrix is used. In the "pca" method, PCA is caculated for these row vectors and the first principal component is used. |
The main matrix and supplementary matrix must have the same columns corresponding to conditions or replicates. They have different features on rows that can be linked by physical location on genomes. The basic assumption is that one feature's variation in the main matrix is correlated with nearby feature(s)' principal variation in the supplementary matrix.
A modified matrix with the same dimensions as the main matrix
data(mancie_example,package="MANCIE") sum_DNase=summarize_mat(exp,ann_exp,DNase,ann_DNase) lev_exp=mancie(exp,sum_DNase)data(mancie_example,package="MANCIE") sum_DNase=summarize_mat(exp,ann_exp,DNase,ann_DNase) lev_exp=mancie(exp,sum_DNase)