Hey Marion,

It can be as simple as creating a combined regression model of the form Disease ~ gene1 + gene2 + ... + gene21 and then deriving test statistics from this, such as AUC, sensitivity, specificity, accuracy, and precision. You will likely reduce this model further —either manually by checking term (gene) p-values, or via other procedures like stepwise regression— as some genes may offer minimal information.

There is a lot of material online, but perhaps no 'all encompassing' tutorial. I have posted various answers that may be of interest to you:

• A: Resources for gene signature creation
• A: How to exclude some of breast cancer subtypes just by looking at gene expressio
• A: What is the best way to combine machine learning algorithms for feature selectio

Any specific questions, please do ask.

Kevin