Microarrays + Neural Networks + Cancer = Coolness

Stumbled on some fresh news showing the potential of microarrays and bioinformatics. Researchers at the National Cancer Institute used a neural network approach (more on this in a bit) coupled to microarray expression profiling to predict the clinical outcome of a conventional treatment against a certain type of cancer (neuroblastoma). Artificial neural networks works in the same way as your neurons. They are pattern recognition algorithms that can 'learn', given a training sample set with known outcomes, how to predict the outcome of new sample sets. In this case, they fed the network with microarrays expression profiles from neuroblastoma patients which had a 'good' (no signs of cancer rebound after 3 years) or 'bad' (died of the disease) future following conventional treatment. The microarrays used quantify more than 25000 genes (and probably ESTs). When the network was sufficiently trained, they fed it with some other samples and were asked to predict the clinical outcome of the treatment (which was known by the scientists, but not by the neural network). 88% accuracy was achieved, which is outstanding. After optimization, they reduced the list of critical genes necessary to correctly predict the outcome to 19. Less genes to predict = less costly and more easy to use in a clinical assay. They are now doing more validation, but this approach could be used in a near future by physicians to give you a more appropriate treatment should you have this type of cancer and predicted to be non-responsive to the conventional therapy. Read about this cool story on News-Medical.net.