aCGH-Smooth Help

Introduction

aCGH-Smooth is a tool for automatic breakpoint identification and smoothing of array-CGH data.

aCGH-Smooth employs a "smoothing" algorithm that tries to adjust the observed array-CGH values such that they represent the copy number of the most common tumor cells. When the observed relative copy numbers (or their logarithms) for the clones are ordered by location on the genome, the values form ``clouds'' with different means, supposedly reflecting different levels of copy numbers, where a change of level represents a breakpoint. The problem is formalized as model fitting to search for most-likely-fit model given the data. A model describes a number of breakpoints, a position for each, and parameters of the distribution of copy number for each. The algorithm estimates the real parameters of the model from the observed array-CGH values.

aCGH-Smooth assumes that the data are generated by a Gaussian process and uses the maximum likelihood criterion for measuring the goodness of a partition, adjusted with a penalization term for taking into account model dimension. A local search procedure searches for a most probable partition of the data using N breakpoints, for a given N.  The procedure is incorporated into a genetic algorithm that evolves a population of partitions with possibly different number of breakpoints that may vary during execution.

aCGH-Smooth is written in visual C++. It has a user friendly interface including a visualization of the results which highlights the obtained smoothing and allows the user to influence the smoothing and number of breakpoints by setting the value of suitable parameters.

In this help we will assume you are familiar with [1].

Requirements

aCGH-Smooth requires Microsoft Windows and Excel 97 or newer installed on your system.

The file format

aCGH-Smooth uses Excel files as input. It can read files produced by the "UCSF Spot" software [2]. The first row should contain the column headers. The following headers should at least be present

• Log2Rat
• Chromosome
• kb

aCGH-Smooth will look for these column headers in the first row of the first sheet of the Excel file. It will look in the first row until it finds a column with an empty header.

aCGH-Smooth will ignore every line in which there is no value for some of these columns. It will stop reading if none of these columns have a value.

There are a few things to note:

• The "Chromosome" column can only contain numbers 1 to 23, so for instance not an "X".
• The "kb" column can only contain numbers, so not for instance ‘nt1234’ (nucleotide).
• aCGH-Smooth adds the ‘kb’ numbers of the last clones for each chromosome, to get an estimation of the total length of the genome (for representation purposes only). This number cannot exceed 2^31 (about 2 billion), so it is important to measure in kilobases and not just basepairs.
• The values in the Excel file should be recognized by Excel as numbers. If Excel does so, it will right align the numbers in the cell. If it does not, a problem may be that you use ‘.’ instead of ‘,’ to mark the decimal point or vice versa.
• Make sure there are no empty column headers before the 3 required column headers.
• There is often a “Flag” column in the file that indicates whether a particular spot contained a successful experiment. This flag is ignored, so you should remove unsuccessful measurements yourself. This can be done by a simple sorting operation on the “Flag” column. There is no need to put the clones back in the right order, because the software will sort the data itself.

Opening files

Opening files can be done in several ways. You can use “open” and “open all” in the file menu or use the icons in the toolbar.

“Open” allows you to select an Excel file. “Open all” allows you to select a directory in which there are some Excel files.

Saving files

Saving files can be done in several ways. You can use “save” and “save all” in the file menu or use the icons in the toolbar.

Saving a file can only be done after smoothing. The dialog window that appears allows you to select 2 options. “Save new ratio” will add an extra column “NewRat” to the Excel file, giving the new found smooth ratio values. “Save type” will add an extra column “Type” to the Excel file, giving the type of each “clone”, that is “loss” (0), “normal” (1), ”gain” (2) or “amplification” (3).

Smoothing files

Smoothing files can be done in several ways. You can use “smooth” and “smooth all” in the edit menu or use the icons in the toolbar. “Smooth” smooths the active or selected document and “smooth all” smooths all opened documents.

Changing parameters

Changing parameters for the smoothing algorithm can be done in several ways. You can use “parameters” in the edit menu or use the icon in the toolbar.

The following parameters are available:

• Lambda: This is the constant for the penalty term for the number of breakpoints in the fitness function of the genetic algorithm, see [1]. Basically increasing it results in finding less breakpoints, and decreasing it results in finding more breakpoints.
• Pool size: This is the number of individuals in the pool of the genetic algorithm.
• Maximum # breakpoints initial pool: This is the maximum number of breakpoints that can occur in individuals of the initial pool. Please note that the genetic algorithm operates per chromosome. So this number of breakpoints is per chromosome.
• Maximum number of generations: This number puts an upper bound on the number of generations for the genetic algorithm.
• Join levels: This check box indicates whether after the smoothing of all chromosomes levels (new ratios) should be joined if they are similar (over all chromosomes). This step makes sense because we expect only a few different levels corresponding to different copy numbers. This step is done using a k-means on the new ratios.
• Maximum # clusters: This is the maximum number of levels that can be found.
• # random starts per fixed # clusters: The algorithm will try to find different numbers of levels and selects the best number of levels. Since k-means is sensitive to the initialization the algorithm will perform a number of random starts of k-means per fixed number of levels. This parameter gives that number of random starts.
• #improvement iterations: For one run of k-means this algorithm gives the number of iterations, that is, the number of times the means are adapted.
• Join closest levels: This option iteratively joins the two closest levels until no pair of levels is closer than some lower bound.
• Minimum difference: This value represents the lower bound mentioned in the previous item.
• Threshold value amplification: If a measurement is above this value we call it an amplification. If the identification of amplifications option is turned on, these measurements will not be considered by the smoothing algorithm.

Looking at data

• The arrow keys can be used to zoom in and out.
• By placing the mouse cursor on a data point, you get information about the corresponding point.

Known issues / bugs

Sometimes if opening a file went wrong, next time you use Excel it does not work. The problem is that after aCGH-Smooth opened a (hidden) instance of Excel, it did not close the Excel application correctly after the error. The temporary solution (at least for Windws NT, 2000 and XP) is to open the Windows Task Manager (right click on taskbar), go to “processes”, locate Excel, right-click Excel and select “end process”.

References

[1] Jong K, Marchiori E, Vaart A van der, Ylstra B, Weiss M, Meijer G. (2003) Chromosomal Breakpoint Detection in Human Cancer. Proceedings EvoBio 2003.

[2] Jain AN, Tokuyasu TA, Snijders AM, Segraves R, Albertson DG, Pinkel D. (2002) Fully automatic quantification of microarray image data. Genome Res. 12:325-32.