I am in pickle when it comes to choose what statistical method I should be using for my data..

At the moment I am working with analyzing clonal heterogeneity in tumors using a new approach. Basically we are using 2D scatterplots to analyze clusters that represent biological events using a visual heuristic approach, like:

enter image description here

As you can see the dots are generally clustering into groups with some outliers, however, this is greatly correlated with quality of data. (The paper from where the figure is from can be found here). I've looked into k-mean cluster analysis, but I don't always know how many clusters we are looking for.

What I had in mind was calculating the x and y mean of each cluster, and calculate some kind of variance/strength. Then compare the same clusters when the data is of lower quality to see how much the variance/strength has increased.

I am unsure what kind of statistics I should be using, mutations are occurring at random, standard error of mean etc. would not work AFAIK. I have looked into Poisson distribution but that is more when trying to calculate the chance of a random event occurring.

Sorry for the long question, I hope you can point me in the right direction! Thanks in advance :)


Clearly a density based approach like DBSCAN and OPTICS will be closest to your intuition.

Nevertheless, I don't think you gain anything by replying on an algorithm here over just continuing to pick clusters yourself. Humans are pretty good at arguing what a pattern is in 2d data - probably better than the algorithms.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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