JS_spliced: Compute the Jensen-Shannon distance between two fitted...

Description Usage Arguments Details Value Author(s) See Also Examples

View source: R/comparative_analysis.R

Description

After models have been fit to your samples, the pairwise JS distance can be computed between them. This function takes the fitted model parameters from 2 distributions and computes the JS distance between them. When all pairwise distances have been computed, they can be used to do hierarchical clustering. This function assumes you denote one distribution as P and one as Q.

Usage

1
2
JS_spliced(grid, shiftp, shiftq, phip, phiq, shapep, shapeq,
           ratep, rateq, threshp, threshq, sigmap, sigmaq, xip, xiq)

Arguments

grid

Vector of integers over which to compute the JS distance. The minimum of the grid is ideally the minimum count of all samples being compared. The maximum is ideally something very large (e.g. 100,000) in order to all or nearly all of the model density. The grid should include every integer in its range. See Examples.

shiftp

The shift for distribution P.

shiftq

The shift for distribution Q.

phip

The estimated phi for distribution P.

phiq

The estimated phi for distribution Q.

shapep

The estimated gamma shape parameter for distribution P.

shapeq

The estimated gamma shape parameter for distribution Q.

ratep

The estimated gamma rate parameter for distribution P.

rateq

The estimated gamma rate parameter for distribution Q.

threshp

The estimated threshold for distribution P.

threshq

The estimated threshold for distribution Q.

sigmap

The estimated parameter sigma for distribution P.

sigmaq

The estimated parameter sigma for distribution Q.

xip

The estimated parameter xi for distribution P.

xiq

The estimated parameter xi for distribution Q.

Details

For 2 discrete distributions P and Q, the Jensen-Shannon distance between them is

JSD(P,Q) = √.5 * [∑(P_i log P_i/M_i)] + ∑(Q_i log Q_i/M_i)

where

M_i= .5 * (P_i + Q_i).

Value

The function directly returns the Jensen-Shannon distance between two fitted distributions P and Q.

Author(s)

hbk5086@psu.edu

See Also

JS_spliced JS_dist

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
data("repertoires")

# Fit the discrete gamma-gpd spliced model at some selected threshold on 2 samples
fit1 <- fdiscgammagpd(repertoires[[1]],
                        useq = quantile(repertoires[[1]], .8),
                        shift = min(repertoires[[1]]))
fit2 <- fdiscgammagpd(repertoires[[2]],
                        useq = quantile(repertoires[[2]], .8),
                        shift = min(repertoires[[2]]))

# Create a grid of every integer from the minimum count to a large value
# The chosen "large value" here is only 1,000, for the sake of quick computation.
# Ideally, the large value will be at least 100,000
grid <- min(c(repertoires[[1]], repertoires[[2]])):1000

# Compute the Jensen-Shannon distance between fit1 and fit2
dist <- JS_spliced(grid,
                    shiftp = min(repertoires[[1]]),
                    shiftq = min(repertoires[[2]]),
                    phip = fit1$mle['phi'], phiq = fit2$mle['phi'],
                    shapep = fit1$mle['shape'], shapeq = fit2$mle['shape'],
                    ratep = fit1$mle['rate'], rateq = fit2$mle['rate'],
                    threshp = fit1$mle['thresh'], threshq = fit2$mle['thresh'],
                    sigmap = fit1$mle['sigma'], sigmaq = fit2$mle['sigma'],
                    xip = fit1$mle['xi'], xiq = fit2$mle['xi'])

dist

powerTCR documentation built on March 17, 2021, 6:01 p.m.