For this deeper analysis, the researchers created an algorithm based on a recently developed computational method designed to recognize individuals by facial features. It is called non-negative matrix factorization, or NMF. In the myeloma study, the algorithm was used to group the results in a way that yielded distinctive genomic features from the CGH data.
Four distinct myeloma subtypes based on genetic patterns emerged: Two of them corresponded to the non-hyperdiploid and hyperdiploid types, and the latter was found to contain two further subdivisions, called k1 and k2 When these subgroups were checked against the records of the patients from whom the samples were taken, it showed that those with the k1 pattern had a longer survival than those with k2.Digging still deeper, the scientists found evidence suggesting that certain molecular signatures within the subgroups are responsible for the differences in outcomes, providing a clear and productive path for further research.
This narrowing down of potential genes and proteins within the subgroups "is a huge advance," comments DePinho. "If you know that a certain gene is driving the disease and influences the clinical behavior of the disease in humans, it immediately goes to the top of the list as a prime candidate for drug development."