0
|
1 library(flashpcaR)
|
|
2
|
|
3 # read in data from either a numeric ped file or an rds object
|
|
4 # output a numeric ped file, with the rownames set to the ids of the samples
|
|
5 get_source_data = function(data_source, data_type) {
|
|
6 data_type = tolower(data_type)
|
|
7 if (data_type == "numeric_ped") {
|
|
8 # read in ped file
|
|
9 ped_data = read.table(data_source, sep="\t", row.names=1)
|
|
10 } else if (data_type == "rds") {
|
|
11 hapmap3_object = readRDS(data_source)
|
|
12 ped_data = hapmap3_object$bed
|
|
13 } else if (data_type == "rdata") {
|
|
14 hapmap3_object = load_obj(data_source)
|
|
15 ped_data = hapmap3_object$bed
|
|
16 } else {
|
|
17 print("Unrecognised data type, returning NULL")
|
|
18 ped_data = NULL
|
|
19 }
|
|
20 return(ped_data)
|
|
21 }
|
|
22
|
|
23 # A function that will read in and return a single object from an RData file
|
|
24 # This is a workaround so the program can run without needing to know name of the object;
|
|
25 # however the assumption is that the RData file contains only ONE object (the one we want)
|
|
26 load_obj = function(filename) {
|
|
27 # create new environment
|
|
28 env = new.env()
|
|
29 # load the rdata file into the new environment, and get the NAME
|
|
30 # of the first object
|
|
31 object_name = load(filename, env)[1]
|
|
32 # return the object
|
|
33 return(env[[object_name]])
|
|
34 }
|
|
35
|
|
36 # remove unwanted rows or columns (samples and snps, respectively) from
|
|
37 # the ped data
|
|
38 filter_ped_data = function(ped_data, xsamples, xsnps) {
|
|
39 # rows to remove
|
|
40 rr = which(rownames(ped_data) %in% xsamples)
|
|
41 # remove rejected samples
|
|
42 if (length(rr) != 0) {
|
|
43 fpd1 = ped_data[-rr, , drop=FALSE]
|
|
44 } else {
|
|
45 fpd1 = ped_data
|
|
46 }
|
|
47 # remove all zero and rejected snp columns
|
|
48 snps = which(colnames(ped_data) %in% xsnps)
|
|
49 zeros = which(colSums(abs(fpd1)) == 0)
|
|
50 cr = union(snps, zeros)
|
|
51 if (length(cr) != 0) {
|
|
52 fpd2 = fpd1[, -cr, drop=FALSE]
|
|
53 } else {
|
|
54 fpd2 = fpd1
|
|
55 }
|
|
56 # remove monomorphic snps
|
|
57 snp_sds = apply(fpd2, 2, sd)
|
|
58 clean_ped = fpd2[, snp_sds >= 0.01, drop=FALSE]
|
|
59 return(clean_ped)
|
|
60 }
|
|
61
|
|
62 # Ethnicity file requirements:
|
|
63 # - tab delimited
|
|
64 # - Must have at least two columns
|
|
65 # - First column has sample ID's
|
|
66 # - Second column has ethnicities
|
|
67 # - First row must be a header
|
|
68 parse_ethnicity_file = function(eth_filename) {
|
|
69 if (file.exists(eth_filename) == FALSE) {
|
|
70 print(paste0("Warning: Ethnicity file: ", eth_filename, " not found"))
|
|
71 return(NULL)
|
|
72 }
|
|
73 if (file.info(eth_filename)$size == 0) {
|
|
74 print(paste0("Warning: Ethnicity file: '", eth_filename, "' is empty"))
|
|
75 return(NULL)
|
|
76 }
|
|
77 eth_data = read.table(eth_filename, header=TRUE, sep="\t")
|
|
78 rownames(eth_data) = eth_data[, 1]
|
|
79 colnames(eth_data)[1] = "IID"
|
|
80 colnames(eth_data)[2] = "population"
|
|
81 return(eth_data)
|
|
82 }
|
|
83
|
|
84 # Read in a file and return the first column as a
|
|
85 # character vector
|
|
86 get_first_column = function(fname) {
|
|
87 rv = c()
|
|
88 if (file.exists(fname) == FALSE) {
|
|
89 print(paste0("Warning: File: '", fname, "' not found"))
|
|
90 return(rv)
|
|
91 }
|
|
92 if (file.info(fname)$size == 0) {
|
|
93 print(paste0("Warning: File: '", fname, "' is empty"))
|
|
94 return(rv)
|
|
95 } else {
|
|
96 rv = as.character(read.table(fname)[, 1])
|
|
97 return(rv)
|
|
98 }
|
|
99 }
|
|
100
|
|
101 # Do pca using flashpcar. Returns a 2 element list
|
|
102 # values - contains the loadings of the pcs
|
|
103 # Will be an n x m matrix, where
|
|
104 # - n = Number of samples
|
|
105 # - m = number of pcs
|
|
106 # ids - Character array of ids, same length as number of rows in values
|
|
107 do_pca = function(ped_data) {
|
|
108 pca_data = list()
|
|
109 pm = data.matrix(ped_data)
|
|
110 values = flashpca(pm, ndim=6)$vectors
|
|
111 pca_data$values = values
|
|
112 pca_data$ids = as.character(rownames(ped_data))
|
|
113 return(pca_data)
|
|
114 } |