Which roles environmental exposures, hereditary factors, and background mutations play in cancer causation is one of the fundamental questions in cancer research. Analyzing their footprints on DNA may enable us to detect and quantify their presence with higher precision than with epidemiological approaches. However, current methods for mutational signatures are based on unsupervised techniques, which have important limitations. In this talk, we present a novel supervised methodology for the detection of mutational signatures in cancer tissues. The mutational signatures obtained by this approach are often very different from their unsupervised counterpart. Importantly, we are able to predict with higher accuracy the presence of exposures and inherited factors, with an average 75% apparent accuracy, and 68% cross-validated, versus a 56% average apparent accuracy using the unsupervised approach. We also present new biological findings obtained thanks to this supervised methodology.