The chromosome database was provided by Alfons Juan and Enrique Vidal (see below), and has 22 classes, with two files for each class, 'difda' and ´difdb´, each with 100 samples (200 total samples per class, 4400 samples in the hole database). The files 'setA' and 'setB' each contain 100 samples of each class and have been used by our group in many experiments, using a 2-fold cross validation scheme.

More information in the message below and also in the IAPR TC5 website.

Francisco Moreno-Seco

gRFIA - Pattern Recognition group

University of Alicante - Spain

This data was kindly provided by Jens Gregor. Here below is his message which describes the database and gives conditions for its use. If you have questions, please ask first to Enrique Vidal (evidal@iti.upv.es).

---------------------------------------------------------------------

From jgregor@cs.utk.edu Wed Oct 2 15:07 MET 1996

Subject: chromosome db intro

Cc: eg@vision.auc.dk

Status: RO

X-Status:

Dear colleagues,

You said you wanted a copy of our chromosome database. Actually, it consists of raw profile data plus a multi-option program for extracting and encoding string sequences. To ensure compatibility, I forward you a copy of the string encodings that we use rather than the raw data itself. For details you will have to see some of the references given below.

We do ask that you make reference to the following paper if, or when, publishing results based on this data:

@article(Lundsteen-al80, author = "C Lundsteen and J Phillip and E Granum", title = "Quantitative analysis of 6985 digitized trypsin {G}-banded human metaphase chromosomes", journal = "Clinical Genetics", volume = 18, pages = "355-370", year = 1980 )

In addition to this reference to the Copenhagen database, as it has become known, you should include one of the following papers (or both) as a reference to the profile processing:

@incollection(Granum-al89, author = "E Granum and M G Thomason and J Gregor", title = "On the use of automatically inferred {M}arkov networks for chromosome analysis", pages = "233--251" editor = "C Lundsteen and J Piper", booktitle = "Automation of Cytogenetics", publisher = "Springer-Verlag", address = "Berlin", year = 1989 )

@article(GraTho90, author = "E Granum and M G Thomason", title = "Automatically inferred {M}arkov network models for classification of chromosomal band pattern structures", journal = "Cytometry", volume = 11, pages = "26--39", year = 1990 )

If you have any questions, I´ll be happy to try and provide an answer. But if the question pertains to the raw data or details of the profile processing, then I suggest that you contact Erik Granum (my Ph.D. advisor). He can be reached as eg@vision.auc.dk. He may also be able to help you if you want to find out about the much larger database that I mentioned (which may or may not exist).

The database consists of 44 files, e.g., dif22da, that each have 100 lines of the form

/ 5467 119 22 27 9 / AA==a==E===d==A==a=Aa=A=a=b

where 5467 is a unique chromosome identifier, 119 refers to the metaphase the sample came from (1..180), 22 is the chromosome type, 27 is the overall string length, and 9 is the length of the p-arm, i.e., the centromere position. The slashes are, of course, only delimiters and should be ignored, i.e., the alphabet consists of the letters a-f, A-F, and = (it may a-e and A-E, I forget).

I will be looking forward to hearing back from you once you have had a chance to apply your methods to the data. As a matter of fact, I´ll be happy find our classification or centromere finding results so that we can make a detailed performance comparison. The papers (cf. the ICGI paper for refs.) only report averages.

Sincerely, Jens Gregor

Files:


dif1da
dif4da
dif2da
dif20da
dif16db
dif17da
dif18da
dif7db
dif4db
dif15db
dif20db
dif10da
dif13db
dif1db
dif5db
dif21db
dif9db
dif14da
dif8da
README
dif16da
dif14db
dif21da
dif5da
dif12db
dif17db
dif22db
dif13da
dif7da
dif6da
dif18db
setB
dif11db
dif11da
dif22da
dif19db
dif3da
dif15da
dif3db
dif10db
dif6db
dif12da
dif8db
dif9da
dif2db
dif19da
setA