Last chance! 50% off unlimited learning
Sale ends in
## S3 method for class 'PredictionProfile,missing':
plot(x, sel = NULL, col = c("red",
"blue"), standardize = TRUE, shades = NULL, legend = "default",
legendPos = "topright", ylim = NULL, xlab = "", ylab = "weight",
lwd.profile = 1, lwd.axis = 1, las = 1, heptads = FALSE,
annotate = FALSE, markOffset = TRUE, windowSize = 1, ...)## S3 method for class 'CrossValidationResult,missing':
plot(x, col = "springgreen")
## S3 method for class 'ModelSelectionResult,missing':
plot(x, sel = c("ACC", "BACC", "MCC",
"AUC"))
## S3 method for class 'ROCData,missing':
plot(x, lwd = 2, aucDigits = 3, cex = 0.8,
side = 1, line = -3, adj = 0.9, ...)
PredictionProfile containing the profiles to be plotted,
for the second method a cross validation result object usually taken
from the trained kebabs model objectFALSE, the profile values s_i
are displayed as they are with the value $y=-b/L$ superimposed as
a light gray line. If TRUE (default), the whole profile is
shifted by $-b/L$ and the light gray line is displayed at y=0.y=-b/L are shaded
in colors shades[1] and shades[2], respectively. Default=NULLlegend is
specified). Can either be a vector with coordinates or a single keyword like
legend).lwd
in parlwd
in parpar"^" below the respective character, for the
lower sequence it is marked by "v" above the sequence. If no offset
element metadata is assigned to the sequences the marks are suppressed.
Default=TRUEInf (infinite)
instead of averages cumulative values along the sequence are used, i.e.
at each position the sum of all contributions up to this position is
displayed. In this case the plot shows how the standardized or
unstandardized value (see parameter standardize) of the
discrimination function builds up along the sequence. Default=1plot command that is called internally to
display the graphics window.parmtextmtextmtextmtextplot method mentioned in the usage section
displays one or two prediction profiles as a step
function with the steps connected by vertical lines. The parameter
sel allows to select the sample(s) if the prediction profile object
contains the profiles of more than two samples. The alignment of the
step functions is impacted by offset metadata assigned to the sequences.
When offset values are assigned one sequence if shifted horizontally to
align the start position 1 pointed to by the offset value for each sequence.
(see also parameter markOffset). If no offset metadata is available
for the sequences both step functions start at their first position on the
left side of the plot. The vertical plot range can be determined by the
rng argument. If the plot is generated for one profile, the sequence
is is visualized above the plot, for two sequences the first sequence is
shown above, the second sequence below the plot. Matching characters at a
position are shown in the same color (by default in "black", the
non-matching characters in the sample-specific colors (see parameter
col). Annotation information can also be visualized along with the
step function. A call with two prediction profiles should facilitate the
comparison of profiles (e.g. wild type versus mutated sequence).The baseline for the step function of a single sample represents the offset b of the model distributed equally to all sequence positions according to the following reformulation of the discriminant function
For standardized plots (see parameter standardize this baseline value
is subtracted from the weight contribution at each position. When sequences
of different length are plotted together only a standardized plot gives
compareable y ranges for both step functions. For sequences of equal length
the visualization can be done in non-standardized or standardized form
showing the lightgray horizontal baseline at positon $y=-b/L$ or at
$y=0$. If the area between the step function and the baseline lying
above the baseline is larger than the area below the baseline the sample
is predicted as belonging to the class assciated with positive values
of the discrimination function, otherwise to the opposite class. (For
multiclass problems prediction profiles can only be generated from the
feature weights related to one of the classifiers in the pairwise or
one-against-rest approaches leaving only two classes for the profile
plot.)
When plotting to a pdf it is recommended to use a height to width ratio
of around 1:(max sequence length/25), e.g. for a maximum sequence length
of 500 bases or amino acids select height=10 and width=200 when opening the
pdf document for plotting.
Plotting of CrossValidation Result
The second variant of plot method shown in the usage section
displays the cross validation result as boxplot.
Plotting of Grid Performance Values
The third variant of plot method shown in the usage section
plots grid performance data as grid with the color of each rectange
corresponding to the preformance value of the grid point.
Plotting of Receiver Operating Characteristics (ROC)
The fourth variant of plot method shown in the usage section
plots the receiver operating characteristics for the given ROC data.
getPredictionProfile,
positionDependentKernel, mcols,
spectrumKernel, mismatchKernel,
gappyPairKernel, motifKernel## set seed for random generator, included here only to make results
## reproducable for this example
set.seed(456)
## load transcription factor binding site data
data(TFBS)
enhancerFB
## select 70\% of the samples for training and the rest for test
train <- sample(1:length(enhancerFB), length(enhancerFB) * 0.7)
test <- c(1:length(enhancerFB))[-train]
## create the kernel object for gappy pair kernel with normalization
gappy <- gappyPairKernel(k=1, m=3)
## show details of kernel object
gappy
## run training with explicit representation
model <- kbsvm(x=enhancerFB[train], y=yFB[train], kernel=gappy,
pkg="LiblineaR", svm="C-svc", cost=80, explicit="yes",
featureWeights="yes")
## compute and plot ROC for test sequences
preddec <- predict(model, enhancerFB[test], predictionType="decision")
rocdata <- computeROCandAUC(preddec, yFB[test], allLabels=unique(yFB))
plot(rocdata)
## generate prediction profile for the first three test sequences
predProf <- getPredictionProfile(enhancerFB, gappy, featureWeights(model),
modelOffset(model), sel=test[1:3])
## show prediction profiles
predProf
## plot prediction profile to pdf
## As sequences are usually very long select a ratio of height to width
## for the pdf which takes care of the maximum sequence length which is
## plotted. Only single or pairs of prediction profiles can be plotted.
## Plot profile for window size 1 (default) and 50. Load package Biobase
## for openPDF
library(Biobase)
pdf(file="PredictionProfile1_w1.pdf", height=10, width=200)
plot(predProf, sel=c(1,3))
dev.off()
openPDF("PredictionProfile1_w1.pdf")
pdf(file="PredictionProfile1_w50.pdf", height=10, width=200)
plot(predProf, sel=c(1,3), windowSize=50)
dev.off()
openPDF("PredictionProfile1_w50.pdf")
pdf(file="PredictionProfile2_w1.pdf", height=10, width=200)
plot(predProf, sel=c(2,3))
dev.off()
openPDF("PredictionProfile2_w1.pdf")
pdf(file="PredictionProfile2_w50.pdf", height=10, width=200)
plot(predProf, sel=c(2,3), windowSize=50)
dev.off()
openPDF("PredictionProfile2_w50.pdf")Run the code above in your browser using DataLab