-1
$\begingroup$

So I have expression data set with 4 healthy sample and 4 disease, to determine the number of K I used mclust which i ran on the list of differentially expressed genes so I got around 7 cluster from Mclust which im using as an input for my K-mean .

Now my data set is like all the differentially expressed genes it contains both UP and down regulated genes so basically it may be UP in my healthy and Down in patient and vice-versa ,4 likely possibilities.

So now I have this doubt that from that I'm getting 7 cluster.Is that result is correct ?

$\endgroup$
1
  • $\begingroup$ What do you mean that you have 4 possibilites? Which constrasts did you do to calculate the differentially expressed genes? Did you compare healthy to disease? Then you only have 2 options. $\endgroup$
    – llrs
    Commented Sep 13, 2018 at 7:45

2 Answers 2

1
$\begingroup$

There is generally no such thing as correct or incorrect when it comes to clustering. When you specify $K$, the program did it's statistical algorithms. Unless the the program had a bug, your results were statistically valid.

You will need to check the clustered labels visually. Only your biological knowledge can tell you the optimal $K$.

$\endgroup$
1
$\begingroup$

7 clusters may be correct in that it's the optimal number according to the Bayesian information criterion (or whatever value you had mclust optimize), but that doesn't make it necessarily the most sensible value. Have a look at your data and visually look at what might make sense. It's likely that the clusters are subdividing up and down regulated genes into various subgroups, which may or may not be useful, depending on your goals.

$\endgroup$
4
  • $\begingroup$ yes I used BIC ,and this is what my PI was asking if those cluster are even correct or make sense ...may be I would try some other clustering and see if get similar number I would let know... $\endgroup$
    – kcm
    Commented Sep 13, 2018 at 10:21
  • 1
    $\begingroup$ As you're learning, clustering tends to be an iterative process. $\endgroup$
    – Devon Ryan
    Commented Sep 13, 2018 at 11:02
  • $\begingroup$ very naive question when i define the number of k , the k mean clustering happens on my rows or columns or both ? $\endgroup$
    – kcm
    Commented Sep 14, 2018 at 5:09
  • 1
    $\begingroup$ Rows are clustered using the values in the columns. $\endgroup$
    – Devon Ryan
    Commented Sep 14, 2018 at 6:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.