Characterizations of the HDMM components on Acute Myeloid Leukemia (AML) from the three different approaches.

Comparison of the three approaches used on a public AML dataset in terms of how components are characterized after the HDMM fit. The left column indicates our multinomial-based approach that is the most rigorous statistical approach, given that the HDMM is run to estimate a mixture of multinomials. In this case, each component is considered as a multinomial and the most frequent genomic alterations are prioritized. The central column reports the driving genomic alterations chosen by the usual standard workflow in onco-hematology. The driver alterations are chosen based on how frequent they are associated with the component and on a priori clinical knowledge. The right column exhibits the prioritization provided when characterizing each component as a MFNCH distribution. This latter seems to show the best compromise between a pure statistical approach (i.e., multinomial-based) and an clinically educated one (i.e., standard workflow). Bold genomic alterations indicate the driving genomic alterations reported by the standard workflow (central column). In both left and right columns only the top six alterations with non-zeros parameters are reported. Plus, beside each alteration the number of times that alteration was clustered by the HDMM into a component is reported. The vertical bar for components 2 and 3 in the central column is a logic OR between drivers, i.e., they are equally prioritized.