| Title: | Iterative Hierarchical Clustering (IHC) |
|---|---|
| Description: | Provides a set of tools to i) identify geographic areas with significant change over time in drug utilization, and ii) characterize common change over time patterns among the time series for multiple geographic areas. For reference, see below: 1. Song, J., Carey, M., Zhu, H., Miao, H., Ram´ırez, J. C., & Wu, H. (2018) <doi:10.1504/IJCBDD.2018.10011910> 2. Wu, S., Wu, H. (2013) <doi:10.1186/1471-2105-14-6> 3. Carey, M., Wu, S., Gan, G. & Wu, H. (2016) <doi:10.1016/j.idm.2016.07.001>. |
| Authors: | Elin Cho [aut, cre], Yuting Xu [aut], Jaejoon Song [aut] |
| Maintainer: | Elin Cho <[email protected]> |
| License: | GNU General Public License (>=3) |
| Version: | 0.1.0 |
| Built: | 2025-03-11 04:01:29 UTC |
| Source: | https://github.com/elincho/ihclust |
This function identifies inhomogeneous clusters using iterative hierarchical clustering (IHC) method.
ihclust( data, smooth = TRUE, cor_criteria = 0.75, max_iteration = 100, verbose = TRUE )ihclust( data, smooth = TRUE, cor_criteria = 0.75, max_iteration = 100, verbose = TRUE )
data |
a numeric matrix, each row representing a time-series and each column representing a time point |
smooth |
if smooth = 'TRUE', a smooth function is applied before clustering |
cor_criteria |
pre-specified correlation criteria |
max_iteration |
maximum number of iterations |
verbose |
if verbose = 'TRUE', the result of a progress is printed |
ihclust
The IHC algorithm implements the three steps as outlined below. First, the Initialization step clusters the data using hierarchical clustering. Second, cluster centers are obtained as an average of all the data points in the cluster. The Merging step considers each of the cluster centers (exemplars) as ‘new data point’, and use the same procedure described in the Initialization step to merge the exemplars into a new set of clusters. Third, the Pruning step streamlines the clusters and removes inconsistencies by reassessing the cluster membership by each data point.
Output from the function is a list of three items:
Cluster_Label - the cluster label for each data point
Num_Iterations - total number of iterations
Unique_Clusters_in_Iteration - unique clusters in each iteration
1. Song, J., Carey, M., Zhu, H., Miao, H., Ram´ırez, J. C., & Wu, H. (2018). Identifying the dynamic gene regulatory network during latent HIV-1 reactivation using high-dimensional ordinary differential equations. International Journal of Computational Biology and Drug Design, 11,135-153. doi: 10.1504/IJCBDD.2018.10011910. 2. Wu, S., & Wu, H. (2013). More powerful significant testing for time course gene expression data using functional principal component analysis approaches. BMC Bioinformatics, 14:6. 3. Carey, M., Wu, S., Gan, G. & Wu, H. (2016). Correlation-based iterative clustering methods for time course data: The identification of temporal gene response modules for influenza infection in humans. Infectious Disease Modeling, 1, 28-39.
# This is an example not using the permutation approach opioid_data_noNA <- opioidData[complete.cases(opioidData), ] #remove NAs mydata <- as.matrix(opioid_data_noNA[1:500,4:18]) testchange_results <- testchange(data=mydata,perm=FALSE,time=seq(1,15,1)) data_change <- testchange_results$sig.change clustering_results <- ihclust(data=data_change, smooth = TRUE, cor_criteria = 0.75, max_iteration = 100, verbose = TRUE)# This is an example not using the permutation approach opioid_data_noNA <- opioidData[complete.cases(opioidData), ] #remove NAs mydata <- as.matrix(opioid_data_noNA[1:500,4:18]) testchange_results <- testchange(data=mydata,perm=FALSE,time=seq(1,15,1)) data_change <- testchange_results$sig.change clustering_results <- ihclust(data=data_change, smooth = TRUE, cor_criteria = 0.75, max_iteration = 100, verbose = TRUE)
A dataset containing estimated opioid dispensing rate per 100 persons in United States, 2006-2020.
data(opioidData)data(opioidData)
data.frame; columns: fips = FIPS county code, State = State, County = County, X2006-X2020 = estimated opioid dispensing rate per 100 persons in each year
https://www.cdc.gov/drugoverdose/rxrate-maps/index.html
This function generates two kinds of datasets. 1. Randomly generates curves with change/no change. 2. Generates true curves assumed from fixed coeffecients with some random noise.
simcurve(numareas = c(300, 300, 300), p = 0.05, type, normerr = 0.1)simcurve(numareas = c(300, 300, 300), p = 0.05, type, normerr = 0.1)
numareas |
number of areas to generate |
p |
proportion of the areas that have significant change |
type |
type of curves generated |
normerr |
standard deviation of the Normal distribution (with mean zero) of which the coefficients are generated |
If type = "random", the function generates curves with change/no change. If type = "fixed", the function generates true curves assumed from fixed coefficients with some random noise. If numareas is not specified, it is assumed as a vector of c(300,300,300). If normerr is not specified, it is assumed as a value of 0.01. It is ignored when type= "random".
Output from the function is a list of two items:
data - simulated data
parameters - parameters used to generate the data
mydata_ran <- simcurve(numareas = c(300, 300, 300), p=0.01, type="random") mydata_fixed <- simcurve(numareas = c(300, 300, 300), p=0.01, type="fixed", normerr = 0.1)mydata_ran <- simcurve(numareas = c(300, 300, 300), p=0.01, type="random") mydata_fixed <- simcurve(numareas = c(300, 300, 300), p=0.01, type="fixed", normerr = 0.1)
This function identifies geographic areas with significant change over time.
testchange(data, time, perm = FALSE, nperm = 100, numclust = 4, topF = 300)testchange(data, time, perm = FALSE, nperm = 100, numclust = 4, topF = 300)
data |
a numeric matrix, each row representing a time-series and each column representing a time point |
time |
defines the time sequence |
perm |
if perm = 'TRUE', a permutation is performed |
nperm |
number of permuations |
numclust |
defines the number of clusters for the parallel processing |
topF |
number of top F values to be selected when perm = 'FALSE' |
number of permutations of >=10,000 is ideal
Output if perm = 'TRUE' is a list of three items:
perm.F - F values obtained from permutation tests
p.values - p-values obtained from permutation tests
p.adjusted - p-values adjusted by Benjamini-Hochberg method
Output if perm = 'False' is a list of three items:
obs.F - conventional F-statistic values
sig.change - areas with significant change over time pattern selected by top F-statistic values
sel.F - top F-statistic values selected
1. Song, J., Carey, M., Zhu, H., Miao, H., Ram´ırez, J. C., & Wu, H. (2018). Identifying the dynamic gene regulatory network during latent HIV-1 reactivation using high-dimensional ordinary differential equations. International Journal of Computational Biology and Drug Design, 11,135-153. doi: 10.1504/IJCBDD.2018.10011910. 2. Wu, S., & Wu, H. (2013). More powerful significant testing for time course gene expression data using functional principal component analysis approaches. BMC Bioinformatics, 14:6. 3. Carey, M., Wu, S., Gan, G. & Wu, H. (2016). Correlation-based iterative clustering methods for time course data: The identification of temporal gene response modules for influenza infection in humans. Infectious Disease Modeling, 1, 28-39.
# This is an example not using the permutation approach opioid_data_noNA <- opioidData[complete.cases(opioidData), ] #remove NAs mydata <- as.matrix(opioid_data_noNA[,4:18]) testchange_results <- testchange(data=mydata,perm=FALSE,time=seq(1,15,1))# This is an example not using the permutation approach opioid_data_noNA <- opioidData[complete.cases(opioidData), ] #remove NAs mydata <- as.matrix(opioid_data_noNA[,4:18]) testchange_results <- testchange(data=mydata,perm=FALSE,time=seq(1,15,1))