Acute Myeloid Leukemia (AML) is a fatal hematological cancer. The genetic abnormalities underlying AML are extremely heterogeneous among patients, making prognosis and treatment selection very difficult. While clinical proteomics data has the potential to improve prognosis accuracy, thus far, the quantitative means to do so have yet to be developed. Here we report the results and insights gained from the DREAM 9 Acute Myeloid Prediction Outcome Prediction Challenge (AML-OPC), a crowdsourcing effort designed to promote the development of quantitative methods for AML prognosis prediction. We identify the most accurate and robust models in predicting patient response to therapy, remission duration, and overall survival. We further investigate patient response to therapy, a clinically actionable prediction, and find that patients that are classified as resistant to therapy are harder to predict than responsive patients across the 31 models submitted to the challenge. The top two performing models, which held a high sensitivity to these patients, substantially utilized the proteomics data to make predictions. Using these models, we also identify which signaling proteins were useful in predicting patient therapeutic response.

Cancer is sometimes depicted as a reversion to single cell behavior in cells adapted to live in a multicellular assembly. If this is the case, one would expect that mutation in cancer disrupts functional mechanisms that suppress cell-level traits detrimental to multicellularity. Such mechanisms should have evolved with or after the emergence of multicellularity. This leads to two related, but distinct hypotheses: 1) Somatic mutations in cancer will occur in genes that are younger than the emergence of multicellularity (1000 million years [MY]); and 2) genes that are frequently mutated in cancer and whose mutations are functionally important for the emergence of the cancer phenotype evolved within the past 1000 million years, and thus would exhibit an age distribution that is skewed to younger genes. In order to investigate these hypotheses we estimated the evolutionary ages of all human genes and then studied the probability of mutation and their biological function in relation to their age and genomic location for both normal germline and cancer contexts.

We observed that under a model of uniform random mutation across the genome, controlled for gene size, genes less than 500 MY were more frequently mutated in both cases. Paradoxically, causal genes, defined in the COSMIC Cancer Gene Census, were depleted in this age group. When we used functional enrichment analysis to explain this unexpected result we discovered that COSMIC genes with recessive disease phenotypes were enriched for DNA repair and cell cycle control. The non-mutated genes in these pathways are orthologous to those underlying stress-induced mutation in bacteria, which results in the clustering of single nucleotide variations. COSMIC genes were less common in regions where the probability of observing mutational clusters is high, although they are approximately 2-fold more likely to harbor mutational clusters compared to other human genes. Our results suggest this ancient mutational response to stress that evolved among prokaryotes was co-opted to maintain diversity in the germline and immune system, while the original phenotype is restored in cancer. Reversion to a stress-induced mutational response is a hallmark of cancer that allows for effectively searching “protected” genome space where genes causally implicated in cancer are located and underlies the high adaptive potential and concomitant therapeutic resistance that is characteristic of cancer.