Entering edit mode
6.9 years ago
arronar
▴
280
Hello out there.
I was thinking to apply decision tree algorithm on my microarrays data as a supervised method for learning but all examples found out there are using binary categorical variable. Either alive or dead, either red or green, either sun or rain and so on. That realization made me wonder if decision trees is a proper algorithm to apply in to microarray data where one can has more than two treatments (i.e Wild type , Medicine 1 , Medicine 2 etc..). What should someone do at this occasion ?
What is your opinion ?
And If I'm wrong, please provide me with some resources that uses tree decision in microarray data (with R).
Thank you.
First of all why only decision tree. You have 10 techniques to be honest and 100's of ML packages. One needs to understand the motivation of the ML methods. Not every ML method will work for your data and then to what extent your test and training model are well fit. If you have enough numbers, it will work but one method might outperform the other depending on the features you are associating. A simple PCA or even regression might work and they are also in the class of ML. There is a very nice workflow for Microarrays which you can try. See this link. Worth giving a try and see how your data behaves and what all features that you want to use gives you a model that is insightful for your hypothesis conclusion. I simply do not believe in using one method picking up from a paper or if someone says it. In cancers when you work with gene expression, not all ML techniques work well and different cancers till date have had different ML methods that fitted well. Especially when meta-analysis was done with them. Obviously, am asking for a bit load of work but it's always worth taking a shot.
You also have to make a decision of the features that you want to use in order to build the hypothesis and what kind of classification are you intending to do. Otherwise, it will be difficult. Take a look at the above link I shared and then play with the data. You might have more to add up once you are comfortable with it.