Hi all,
I am running a Random Forest –Mean Decrease in Accuracy algorithm for feature selection on my Microarray data in order to use the selected genes as a classifier to discriminate between 2 classes of cell lines. I am having problems to interpret the output information given by the algorithm. It gives me a small list of selected genes and for each gene there is a Pearson correlation value, a fold change value and a q-value (False Discovery Rate) .
The variable “class” is discrete (normal vs disease), so what does the Pearson correlation mean in this case?
Should I take the q-value showed as a multiple test correction and give less importance, or exclude, the genes that showed a q-value (FDR) higher than 0.2 (or any other pre-determined cut-off for significance)?
I would appreciate any suggestion on how to interpret the results of a RF-MDA for feature selection algorithm.
Hi Priscila,
What software are you using to run RF-MDA test? I suggest you to check the software manual to see if there is any explanation and/or example regarding this topic.
Moreover, you can check the following link, you may find some insights:
https://stats.stackexchange.com/questions/197827/how-to-interpret-mean-decrease-in-accuracy-and-mean-decrease-gini-in-random-fore
Best regards,
Hi Eduardo,
Thank you very much for the link! I am using a online GUI platform to run the algorithm in R language, I could not find such a manual.
Best wishes,
Dearest,
to your question : " Should I take the q-value showed as a multiple test correction and give less importance, or exclude, the genes that showed a q-value (FDR) higher than 0.2 (or any other pre-determined cut-off for significance)? ", and with the given information. I think it will be more usefull to analyze the specific group on correlation effectivness, and searching the "why". They go ahaed. We are too oriented in laerning process in Europe not open enough and learn to exclude but the originality will be to undertaken the necessary action to validate or not a new concept based on an other singificant cutt-off. I will exclude after testing and assement. Sincerely Yours
Hello Priscila,
Since you mentioned about using a GUI and not having a manual, you may try contacting the developers for the best explanation (possibly may have a look at the source code), to be exactly sure what the pearson coeffs and q values related to. In the mean time, you may have a look at the following studies where they have explained the importance of various correlation factors in RF-MDA:
Article An experimental study of the intrinsic stability of random f...
https://blog.datadive.net/selecting-good-features-part-iii-random-forests/
Thanks
Hi Sayeed,
Good ideia, I will contact the developers! Thank you very much for the literature recomendation!
Best,
- Badges
- Science topic