Avro - I would suggest - keep reading! If you are interested in historical perspective, this paper was among the first to demonstrate useful classification of cancers by gene expression, using a very simple method. Since that time many other methods have been applied. Often, there are several different supervised methods that can work reasonably well for a given dataset. For example, in a recent case we worked on, techniques like CART, linear discriminants, and SVNs gave similar performance. Your mileage may vary, depending on your scenario. It's important to focus on fundamentals, such as feature selection, avoiding over-fitting, especially on datasets with low numbers of samples, and always validating trained predictors on independent test datasets.