,
Anthony Gitter [aut, cph] | Title: | Lag Penalized Weighted Correlation for Time Series Clustering |
|---|---|
| Description: | Computes a time series distance measure for clustering based on weighted correlation and introduction of lags. The lags capture delayed responses in a time series dataset. The timepoints must be specified. T. Chandereng, A. Gitter (2020) <doi:10.1186/s12859-019-3324-1>. |
| Authors: | Thevaa Chandereng [aut, cre, cph]
,
Anthony Gitter [aut, cph] |
| Maintainer: | Thevaa Chandereng <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 1.0.0 |
| Built: | 2025-02-10 05:14:56 UTC |
| Source: | https://github.com/gitter-lab/lpwc |
best.lag computes the best lags for a dataset using weighted correlation.
The lags obtained are in reference to the original timepoints.
best.lag(data, timepoints, max.lag = NULL, C)best.lag(data, timepoints, max.lag = NULL, C)
data |
a matrix or data frame with rows representing genes and columns
representing different timepoints. If data is a data frame, the gene names
can be specified using the |
timepoints |
a vector of time points used in the dataset |
max.lag |
a numeric value of the maximum lag allowed, if null, defaults to the floor of the number of timepoints divided by 4 |
C |
a numeric value of C used in computing weighted correlation |
a vector of best lags used in the dataset, one per gene
Thevaa Chandereng, Anthony Gitter
best.lag(data = array(rnorm(20), c(4, 5)), timepoints = c(0, 5, 10, 20, 40), C = 300) best.lag(data = array(runif(100), c(5, 20)), timepoints = seq(2, 40, 2), C = 10) best.lag(data = array(runif(100), c(5, 20)), timepoints = seq(2, 40, 2), max.lag = 2, C = 10)best.lag(data = array(rnorm(20), c(4, 5)), timepoints = c(0, 5, 10, 20, 40), C = 300) best.lag(data = array(runif(100), c(5, 20)), timepoints = seq(2, 40, 2), C = 10) best.lag(data = array(runif(100), c(5, 20)), timepoints = seq(2, 40, 2), max.lag = 2, C = 10)
This function computes the weighted correlation with a penalty for lags. It should only be used after the fixed lags have already been applied to the dataset and timepoints using the functions prep.data() and best.lag().
comp.corr(data, time, C)comp.corr(data, time, C)
data |
a lagged matrix or data frame with rows representing genes and columns representing different timepoints (NAs added when lags are needed) |
time |
a lagged matrix with rows representing each gene's timepoint and columns representing the number of timepoints, NA is introduced when it is lagged |
C |
a numeric value of C used in computing weighted correlation |
a simmilarity matrix with values between -1 and 1 (1 highly correlated, 0 no correlation)
Thevaa Chandereng, Anthony Gitter
## This function computes the correlation after the lags (or shifts) have ## been computed. In this example, the lags argument is randomly sampled ## for the sake of illustrating how prep.data() applies the lags and ## prepares a transformed dataset for comp.corr(). lagged <- prep.data(array(rnorm(30), c(3, 10)), timepoints = seq(0, 45, 5), lags = sample(c(0, 1, -1, 2, -2), size = 3)) comp.corr(data = lagged$data, time = lagged$time, C = 10) ## This example shows how comp.corr is used in practice with real data. ## The best.lag() function is called first to pre-compute the lags, which ## are passed to prep.data(). randdata <- array(rnorm(120), c(10, 12)) bl <- best.lag(data = randdata, timepoints = 1:12, C = 5) lag.data <- prep.data(randdata, timepoints = 1:12, lags = bl) comp.corr(lag.data$data, time = lag.data$time, C = 5)## This function computes the correlation after the lags (or shifts) have ## been computed. In this example, the lags argument is randomly sampled ## for the sake of illustrating how prep.data() applies the lags and ## prepares a transformed dataset for comp.corr(). lagged <- prep.data(array(rnorm(30), c(3, 10)), timepoints = seq(0, 45, 5), lags = sample(c(0, 1, -1, 2, -2), size = 3)) comp.corr(data = lagged$data, time = lagged$time, C = 10) ## This example shows how comp.corr is used in practice with real data. ## The best.lag() function is called first to pre-compute the lags, which ## are passed to prep.data(). randdata <- array(rnorm(120), c(10, 12)) bl <- best.lag(data = randdata, timepoints = 1:12, C = 5) lag.data <- prep.data(randdata, timepoints = 1:12, lags = bl) comp.corr(lag.data$data, time = lag.data$time, C = 5)
This function computes correlation based on best picked lags. The lags indicate delayed changes.
corr.bestlag(data, timepoints, max.lag = NULL, C = NULL, penalty = "high", iter = 10)corr.bestlag(data, timepoints, max.lag = NULL, C = NULL, penalty = "high", iter = 10)
data |
a matrix or data frame with rows representing genes and columns
representing different timepoints. If data is a data frame, the gene names
can be specified using the |
timepoints |
a vector of time points used in the dataset |
max.lag |
a integer value of the maximum lags allowed in the dataset, if null, defaults to the floor of the number of timepoints divided by 4 |
C |
a numeric value of C used in computing weighted correlation, if null, a default is computed based on the penalty argument |
penalty |
a factor with two levels high and low penalty on the weighted correlation |
iter |
an integer indicating the number of C values to test for low penalty |
a list containing weighted correlation and best lags used in each row
Thevaa Chandereng, Anthony Gitter
corr.bestlag(array(rnorm(30), c(5, 6)), max.lag = 1, timepoints = c(0, 5, 10, 15, 20, 25), C = 10, penalty = "high") corr.bestlag(array(runif(40, 0, 20), c(4, 10)), timepoints = c(0, 0.5, 1.5, 3, 6, 12, 18, 26, 39, 50), penalty = "high") corr.bestlag(matrix(data = rexp(n = 40, 2), nrow = 8), timepoints = c(0, 5, 15, 20, 40), penalty = "low", iter = 5)corr.bestlag(array(rnorm(30), c(5, 6)), max.lag = 1, timepoints = c(0, 5, 10, 15, 20, 25), C = 10, penalty = "high") corr.bestlag(array(runif(40, 0, 20), c(4, 10)), timepoints = c(0, 0.5, 1.5, 3, 6, 12, 18, 26, 39, 50), penalty = "high") corr.bestlag(matrix(data = rexp(n = 40, 2), nrow = 8), timepoints = c(0, 5, 15, 20, 40), penalty = "low", iter = 5)
This function computes the values of C to test using the timepoints and max lag in the dataset
findC(timepoints, max.lag = NULL, pi = 0.95, iter = 10)findC(timepoints, max.lag = NULL, pi = 0.95, iter = 10)
timepoints |
a vector of timepoints used in the dataset |
max.lag |
a numeric value with maximum lags allowed, if null, defaults to the floor of the number of timepoints divided by 4 |
pi |
a numeric value between 0.5 and 1 for the upper bound on the penalty |
iter |
a numeric value with the number of penalties to test |
a vector of length iter of the different values of C to test
Thevaa Chandereng, Anthony Gitter
findC(c(0, 5, 10, 15, 20, 25), max.lag = 1, iter = 15) findC(c(2, 4, 8, 16, 32, 64, 128, 256), iter = 5) findC(c(2, 6, 10, 15, 22, 30, 40, 55, 80), pi = 0.8, iter = 20) findC(c(1, 2, 3.2, 4, 5.3, 7), pi = 0.99)findC(c(0, 5, 10, 15, 20, 25), max.lag = 1, iter = 15) findC(c(2, 4, 8, 16, 32, 64, 128, 256), iter = 5) findC(c(2, 6, 10, 15, 22, 30, 40, 55, 80), pi = 0.8, iter = 20) findC(c(1, 2, 3.2, 4, 5.3, 7), pi = 0.99)
This function prepares the data to compute correlation by introducing NA's when lags are needed
prep.data(data, lags, timepoints)prep.data(data, lags, timepoints)
data |
a matrix or data frame with rows representing genes and columns
representing different timepoints. If data is a data frame, the gene names
can be specified using the |
lags |
a vector of same length as the number of rows in the data column indicating the best lags |
timepoints |
a vector of time points used in the dataset |
a list of two matrices, one matrix with NA's for the lags for the dataset and another matrix with the timepoints used for each row in the dataset
Thevaa Chandereng, Anthony Gitter
prep.data(array(rnorm(20), c(5, 4)), c(0, 0, 0, -1, 1), timepoints = c(0, 5, 15, 30)) prep.data(array(runif(100, 0, 10), c(10, 10)), sample((-2:2), size = 10, replace = TRUE), timepoints = c(0, 5, 15, 30, 45, 60, 75, 80, 100, 120))prep.data(array(rnorm(20), c(5, 4)), c(0, 0, 0, -1, 1), timepoints = c(0, 5, 15, 30)) prep.data(array(runif(100, 0, 10), c(10, 10)), sample((-2:2), size = 10, replace = TRUE), timepoints = c(0, 5, 15, 30, 45, 60, 75, 80, 100, 120))
This function computes the score of best lags by summing the correlation of corresponding lags
score(corr, lags)score(corr, lags)
corr |
a vector of computed correlation |
lags |
a vector of same length with corr that holds the lags corresponding to the corr vector |
a numerical value of best lag picked based on corr and lags
Thevaa Chandereng, Anthony Gitter
score(runif(10, 0, 1), c(2, 0, 0, 0, 3, 2, -1, 2, 0, 1)) score(runif(20, 0.5, 0.8), sample(-3:3, size = 20, replace = TRUE))score(runif(10, 0, 1), c(2, 0, 0, 0, 3, 2, -1, 2, 0, 1)) score(runif(20, 0.5, 0.8), sample(-3:3, size = 20, replace = TRUE))
Data is simulated from ImpulseDE function with 8 timepoints c(0, 2, 4, 6, 8, 18, 24, 32, 48, 72) See https://doi.org/10.1093/bioinformatics/btw665
simdatasimdata
data matrix
data(simdata)data(simdata)
This function computes the weight used for correlation using timepoints used and lags used
weight(t, lag, C)weight(t, lag, C)
t |
a vector of timepoints |
lag |
a integer value of the lag used |
C |
a numeric of the constant used in the penalty and weight inside the Gaussian kernel |
a list containing w0 and vector w used for computing weighted correlation
Thevaa Chandereng, Anthony Gitter
weight(t = c(0, 5, 10, 15, 20), lag = 1, C = 20) weight(t = c(0, 2, 5, 10, 14, 19, 22), lag = 1,C = 100)weight(t = c(0, 5, 10, 15, 20), lag = 1, C = 20) weight(t = c(0, 2, 5, 10, 14, 19, 22), lag = 1,C = 100)
This function matches vectors of two different lengths
weight.lag(x1, x2)weight.lag(x1, x2)
x1 |
a vector |
x2 |
a vector |
a matrix with two rows with the shortest length of the vector as the number of columns
Thevaa Chandereng, Anthony Gitter
weight.lag(1:5, 2:9) weight.lag (seq(0, 10, 2), seq(4, 10, 2))weight.lag(1:5, 2:9) weight.lag (seq(0, 10, 2), seq(4, 10, 2))
This function computes weighted Pearson correlation between two vectors with weights given. The output is between -1 and 1 with 1 being highly positively correlated, -1 being highly negatively correlated, and 0 being no correlation
wt.corr(x, y, w)wt.corr(x, y, w)
x |
a vector |
y |
a vector with same length of x |
w |
a vector with same length of x and y |
a numerical value of weighted Pearson correlation
https://en.wikipedia.org/wiki/Pearson_correlation_coefficient#Weighted_correlation_coefficient
wt.corr(c(1, 2, -9, 4, 5), c(2:6), c(0.5, 1, 2, 0.5, 2)) wt.corr(rnorm(5), rnorm(5), runif(5, 0, 1))wt.corr(c(1, 2, -9, 4, 5), c(2:6), c(0.5, 1, 2, 0.5, 2)) wt.corr(rnorm(5), rnorm(5), runif(5, 0, 1))