inconsistent results

[Petros Kolovos]

Dear All,

I have a chip-seq and I comparing that vs another chip-seq (reference)
and everytime that I do the clustering (Kmeans raw) with 10
clusters.....everytime the picture in the end is different than the
previous time

why is this happening?

thank you in advance

best regards
Petros

[Petros Kolovos]

Dear All,

here you can see some examples of my problem

All the 3 pictures were created from the same dataset with 10 clusters WITHOUT changing any parameter

I understand that the problem is that the program takes everytime different clusters

Isn't a way to define how the clusters are going to be taken? For example to define one cluster that is peaks that >1000 bp from TSS, another cluster that is peaks <1000 bp and so on.....

Who this can be done?

Thank you in advance

I will wait for your asnwer

Yours sincerely
Petros Kolovos

2012/5/21 Petros Kolovos <p.ko...@gmail.com>

[Tao]

Hi Petros,

I'm afraid that you've mis-understood the k-means clustering method.

you will have a reference here to explain your questions.
http://en.wikipedia.org/wiki/K-means_clustering

best,
Tao

On May 23, 5:04 pm, Petros Kolovos <p.kolo...@gmail.com> wrote:
> Dear All,
>
> here you can see some examples of my problem
>
> All the 3 pictures were created from the same dataset with 10 clusters
> WITHOUT changing any parameter
>
> I understand that the problem is that the program takes everytime different
> clusters
>
> Isn't a way to define how the clusters are going to be taken? For example
> to define one cluster that is peaks that >1000 bp from TSS, another cluster
> that is peaks <1000 bp and so on.....
>
> Who this can be done?
>
> Thank you in advance
>
> I will wait for your asnwer
>
> Yours sincerely
> Petros Kolovos
>

> 2012/5/21 Petros Kolovos <p.kolo...@gmail.com>

>
>
>
>
>
>
>
> > Dear All,
>
> > I have a chip-seq and I comparing that vs another chip-seq (reference)
> > and everytime that I do the clustering (Kmeans raw) with 10
> > clusters.....everytime the picture in the end is different than the
> > previous time
>
> > why is this happening?
>
> > thank you in advance
>
> > best regards
> > Petros
>
>
>

>  1.png
> 139KViewDownload
>
>  2.png
> 138KViewDownload
>
>  3.png
> 139KViewDownload

[Petros Kolovos]

Dear Tao,

thank you very much. I understand what K-mean means

So, if i understand correct....whatever normalization I will use, if I repeat the analysis I will always have different results?

That means that you have to repeat your analysis many times in order to have a picture which fits to your expectation?

Isn't scarry that you have different pictures each time you do the analysis with the same input?

Isn't a way to define the way the clusters are made? Like I said in the previous email, for example to have at one cluster the peaks which have a specific distance from TSS (for example greater that 1000bp from TSS) and another cluster to have peaks with distance less than 1000bp from TSS.

So if i want to see the distribution of my peaks relevant to TSS, is it logical to repeat the clustering method many times until I have the best picture? How can I define what number is the best to choose for the number of clusters?

Thank you in advance

I will wait for your answer as soon as possible as I really have to make some pictures with seqminner

Best regards
Petros

2012/5/28 Tao <ytsq...@gmail.com>

[Gema Sanz]

Hi Petros, if you use a fixed k-mean seed you will get always the same result (toolsoptions/general/run kmeans with a given seed, I use 5)