Where To Find Test Datasets For Data Classification Problems

Question

[emanuele.pascariello]

HI there.

i am new to this blog and bioinfo too.

i am finishing my master degree in bioinformatics and findign very interesting the data mining field. I have started to practice and perform some analysis on Acute Lymphoblastic Leukemya and acute myeloid datasets based on the famous Golub dataset from here http://datam.i2r.a-star.edu.sg/datasets/krbd/Leukemia/ALLAML.html Hi understand it is one of the most famous and studied. Almost every paper talks about it.

My question is, once i finished my analysys on this dataset in order to validate these conclusion i would like to challenge the methods with some other dataset related to the same classification problem (ALL; AML). Any suggestion where can i get some different dataset?

Thanks

[Sean Davis]

It sounds like you are pretty new to bioinformatics. I'd suggest finding a local collaborator to help you navigate the database and format details until you are a bit more comfortable with the data. It will certainly save you some time and energy in the short term and you will learn more in the long term.

[emanuele.pascariello]

yes i am..as i said before. Anyway thanks for the tips i am starting to figure out all this stuff.

Thanks

[Neilfws]

If you have further questions, please can you post them as comments under each answer, not as new answers.

Answer 1

[Sean Davis]

The largest collection of publicly available microarray data are at NCBI GEO. A large collection of cancer samples is available in GSE2109, but there are literally hundreds of potentially interesting datasets there.

Answer 2

[emanuele.pascariello]

That was very helpful what is the difference among GPL and GSE accession?

Thanks

[Sean Davis]

http://www.ncbi.nlm.nih.gov/geo/info/overview#org

[emanuele.pascariello]

ok.... Thanks for your tip Regards

Answer 3

[emanuele.pascariello]

sorry for bothering you again, there is something is not clear to me.

i have always worked with arff files already formatted, with several instances thousands of attributes and divided in to classes. If i want to try another dataset (with many patients, instances) i have to obtain an arff file containing several intances (1 instance==1 patient) since i am using weka as a data mining software. My question: is this http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM353943 (let's pretend this is the case i want) the expression array related to just one patient? what if i want many of these arrays (for many patients) and then want to obtain a single arff file divided in to classes in order to work with weka? I probably may seem very naive to you and let's say i am indeed, but i am at the very beginning of this.

Thanks

[Sean Davis]

Yes, GSMs represent single samples. GSE and GDS records represent collections of related samples. I'm not sure what you mean by "arff" format. The format used by NCBI GEO is SOFT format or MiNIML, both of which are specific to NCBI GEO.

[emanuele.pascariello]

arff is the input file for weka data mining software, Attribute-Relation File Format.

[bgupta.rs.cse]

pl give ur mail id, i need to talk to u for neuropsychiatric disease protein or snp dataset in arff format. my mail [e-mail],[phone]

[Sean Davis]

Ahh, I see. You can simply convert the microarray data into arff format yourself. Alternatively, if you want to pursue machine learning, you could use R directly since the GEOquery package loads data from NCBI GEO directly into R.

Answer 4

[mattia.bosio85]

You can always look for the cancer genome atlas datasets tcga-data.nci.nih.gov/tcga they have an acutemyeloid leukemia dataset among others. I don't know about ALL samples instead.

Answer 5

[emanuele.pascariello]

Thanks a lot to all. still trying to figure out this matter. anyway it's fascinating Regards

Answer 6

[Neilfws]

There was a recent blog post on this topic: "High-Dimensional Microarray Data Sets in R for Machine Learning." The author has created an R package containing 20 or so datasets.

[emanuele.pascariello]

Thank you, i'll give it a look asap.