I am currently working on a data set regarding Sarcoma cancer. My data set contains the regions that are defected in patients’ chromosomes. I allocated one to defected regions and zero to healthy ones.
So, my data for each patient is like a long string of 1 and 0. I am supposed to classify each patient disease subtype and risk group according to labels I have. I have data for 44 patients and so for the method that have come to my mind is SVM with kernel methods.
If you have any suggestion or recommendation for applying another method, I would be more than happy to know it.
Hi Nastaran,
This seems to be an epidemiological descriptive task.
You can make subgroups yourself, and test them for each variable with chi2 test/student's t test.
If you are looking for a program that does subgroups for you, I would suggest going with log regression. It works well with such types.
You can also do a Tree classification, generating a tree. It will give you more simple readable format.
Exploring with hierarchical clustering can be nice as well with the first branches to be the outcome.
A good way to start is to create a PCA of your variables. Then display outcome as two clusters. You'll likely to see if patients in the clusters are distinctive with each other.
To test statistical significance, all methods are valid. it just depends what you are looking for... either risk factors (log regression), subgroups (Tree, Hierarchical).
Log regression tests significance.
To test subgroups in Tree and hierarchical, you still need to use classical statistics to test de significance.
All of this is retrospective and with few patients, you can only do descriptive analytics.
Yohann
SVM tends to give better result when you work with many features.
Dear Yohann Gabriel Missiak
Thank you for your comprehensive reply. I will update you about the outcome soon.
Syed A. Taqvi
Pritom Kumar Saha Thank you for your reply
You are using supervised machine learning algorithm by assigning labels to data. In addition to SVM there are others algorithm that can be used such as K-nearest neighbors (KNN), decision tree (DT), and linear Discriminant algorithm (LDA).
Few related research articles are:
Article Machine learning applications in cancer prognosis and prediction
The above article discussed use of Artificial Neural Networks (ANNs), Bayesian Networks (BNs), Support Vector Machines (SVMs) and Decision Trees (DTs) in cancer research for the development of predictive models
Article Applications of Support Vector Machine (SVM) Learning in Can...
Hi Murtaza Khan ,
Thank you for your reply.
Actually my concern is how to deal with my data. As I explaind it would be long strings of 0 and 1.
Hi Nastaran Maus Esfahani,
For data in the vectorial forms, you can use both classical methods and deep learning. However, I do not suggest deep learning because you have only more than 40 examples. As you and others said, we can use kernel method (SVM is an example). Since each example in your dataset is represented by a binary vector, I would suggest you to use boolean kernel with SVM. For measure the performance, you can use leave one out to have more examples in the training set.
- Badges
- Science topic