Our three preferred algorithms are
(1) Circular Binary Segmentation
(Olshen et al, /Biostatistics, /2004)
(2) Gaussian model with adaptive penalty
(Picard et al, /BMC Bioinformatics, /2005) and
(3) Fused Lasso
(Tibshirani and Wang, /Biostatistics /2007).
The first two were chosen based on their excellent performance
on our comparative study (Lai etal, /Bioinformatics, /2005),
the third is a more recent method that appears to work well.

BioHMM uses a heterogeneous hidden Markov model. By default, it considers the distance between probes when estimating its parameters and gives higher probabilities to probes that are further apart than others. This algorithm is called from the snapCGH package, which is available in BioConductor.

Parameters:
Use clone distances: tells the algorithm either to consider probe distances in its calculations or assume a homogeneous hidden Markov model instead.Enabled by default.

Citation:Marioni, J.C., Thorne, N.P., and Tavare, S. (2006). BioHMM: a heterogeneous hidden Markov model for segmenting array CGH data. Bioinformatics 22:1144-6.

CBS estimates the location of change-points by calculating a likelihood-ratio statistic for each probe and assessing its significance by permutation. This algorithm is called from the DNAcopy package, which is available in BioConductor.

Parameters:
Alpha: the likelihood by chance that the segment means surrounding the change-point are equal.
Default: 0.05 Range: min > 0.0 max < 1.0

Citation: Olshen, A.B., Venkatraman, E.S., Lucito, R., and Wigler, M. (2004). Circular binary segmentation for the analysis of array-based dna copy number data. Biostatistics 5:557-72

FASeg uses lowess to find the location of possible breakpoints and conducts local ANOVA to identify significant breakpoints. This algorithm is called from the FASeg package, which is available from http://www.sph.emory.edu/bios/FASeg/

Parameters:
sig: Significance cutoff value.
Default: 0.025 Range: min > 0.0 max < 1.0
delta: The minimum height of the "bumps" in the lowess-smoothed data to consider their boundaries as potential breakpoints
Default: 0.1 Range: min > 0.0 max 0.5
smooth range: the number of probes to use when calculating the weights around each probe in the lowess smoother.
Default: 50 Range: min 10 max 100
Fine Tune:tells the algorithm to recalculate breakpoint locations (in smaller neighborhoods) after edge selection.
Disabled by default.

Citation: Yu, T., Ye, H., Sun, W., Li, K., Chen, Z., Jacobs, S., Bailey, D., Wong, D.T., Zhou, X. (2007). A forward-backward fragment assembling algorithm for the identification of genomic amplification and deletion breakpoints using high-density single nucleotide polymorphism (SNP) array. BMC Bioinformatics 8:145.

cghFLasso smoothes the data with the fused lasso, a spatial smoothing technique. This algorithm is called from the cghFLasso package, which is available from http://www-stat.stanford.edu/~tibs/cghFLasso.html. Because of cghFLasso's memory requirements, this website will divide the chromosome into smaller pieces if it has more than 10000 probes.

Parameters:
FDR: False discovery rate (the proportion of true null hypotheses among those called significant).
Default: 0.05 Range: min > 0.0 max < 1.0
Use this value: tells the algorithm to use FDR to determine significant segments.
Disabled by default
Recalculate Segment Means: A post-processing step (not part of cghFLasso) to recalculate the segment means after finding the breakpoints with cghFLasso.
Disabled by default.

Citation: Tibshirani, R., Wang, P. (2007). Spatial smoothing and hot spot detection for CGH data using the fused lasso. Biostatistics [Epub ahead of print].

CGHseg estimates breakpoints by making a cost matrix, finding all possible breakpoints from this matrix, and selecting the most likely number of breakpoints with adaptive penalty. The algorithm has been rewritten in C based on the MATLAB code provided by Picard et al. on their website (http://www.inapg.fr/ens_rech/maths/outil_A.html). Because of CGHseg's memory requirements, this website will divide the chromosome into smaller pieces if it has more than 10000 probes..

Parameters:
Km:The maximum number of segments to consider per chromosome.
Default: 20 Range: min 5    max 50
S: The adaptive penalty threshold.
Default: -0.5 Range: min -1.0   max < 0.0

Citation: Picard, F., Robin, S., Lavielle, M., Vaisse, C., Daudin, J.J. (2005). A statistical approach for array CGH data analysis. BMC Bioinformatics 6:27.

GLAD smoothes the data with likelihood-based adaptive weights smoothing, removes extraneous breakpoints with a penalized likelihood, and groups the segments with unsupervised clustering. This algorithm is called from the GLAD package, which is available in BioConductor.

Parameters:
qlambda: a scaling parameter used by adaptive weights smoothing for its stochastic penalty
Default: 0.9990 Range: min 0.9000 max 0.9999

Citation: Hupe, P., Stransky, N., Thiery, J.P., Radvanyi, F., Barillot, E. (2004). Analysis of array CGH data: from signal ratio to gain and loss of DNA regions. Bioinformatics 20(18):3413-22

LOWESS smoothes the data with robust weighted local polynomial fitting. Probes inside the smoothing window are weighted according to their distance from the center, with the more distant probes having less weight. This algorithm is called from the stats package of R.

Parameters:
Width: the number of probes to use when calculating the weights around each probe in the lowess smoother.
Default: 15 Range: min 5 max 50

Wavelet smoothing smoothes the data by transforming the data into frequency components with maximal overlap discrete wavelet transform. The transformed data are filtered through soft SURE thresholding and then transformed back to the time domain to get the smoothed data. This approach is similar to the procedure described in Hsu et al. (2005). This algorithm uses functions from the waveslim package, which is available in the Comprehensive R Archive Network.

Parameters:
Levels: the depth of decomposition for maximal overlap discrete wavelet transform and the depth of thresholding for SURE.
Default: 3 Range: min 1 max 6

Quantile smoothing uses penalized quantile regression to find trends in the data. The code follows closely to the R code outlined in the Eilers and de Menezes (2005) paper, except we use the sparse implementation of the Frisch-Newton interior-point algorithm. The results represent the 50th quantile. This algorithm uses functions from the quantreg package, which is available in the Comprehensive R Archive Network.

Parameters
Lambda: The penalty parameter for the regression.
Default: 2.0 Range: min > 0.0 max 10.0

Citation:Eilers, P.H.C. and de Menezes, R.X. (2005). Quantile smoothing of array CGH data. Bioinformatics 21:1146-1153.

This method takes the average of probe values inside a smoothing window. The code is written in C.

Parameters
Width: The number of probes to use around a probe when calculating their means
Default: 15 Range: min 5 max 50

Citation:NONE

These parameters adjust the minimum and maximum log-ratios to show in the graphical results. Log-ratios below the minimum are drawn at the minimum value. Log-ratios above the maximum are drawn at the maximum value.

Parameters:
Minimum Log-Ratio:
Default: -2.0 Range: min -1.0 max -10.0
Maximum Log-Ratio:
Default: 2.0 Range: min 1.0 max 10.0

This is a simple way to identify gains and losses in the processed data. Gains are called for regions above the positive threshold. Losses are called for regions below the negative threshold.