Acute myeloid leukemia (AML) is an aggressive blood cancer with dismal survival rates and extreme variability from patient to patient. This poses a serious challenge to the development of effective new therapies.
After a decades-long drought, the Food and Drug Administration (FDA) approved four new drugs for the disease in 2017. One of these, Idhifa (enasidenib), resulted directly from discoveries in major genomic profiling projects. A transformative drug was brought to bear in some of the worst-prognosis cases of the disease, by combining functional studies with insight from data-intensive molecular profiling. A New England Journal of Medicine editorial heralded the first of these studies as “the beginning of the end of the beginning”—as a chapter closed on pure profiling studies, a new chapter has opened, where large-scale clinical trials are wedded to deep characterization of patients and their responses.
In this new chapter, perhaps best exemplified by the genetic diversity of leukemia both between and within patients, there is no one cure for cancer—there are thousands. As novel agents prove their worth, a new problem is emerging: how can we couple the statistical rigor of randomized trials—a foundational tool of modern medicine—with the recognition that what we have long treated as a single disease in fact contains dozens?
The Triche Laboratory is focused on reconciling these opposing forces. Quick-to-fail early trial designs, enriched for specific patients with a high probability of response, can hasten the real-world testing of new drugs, which in turn are prioritized by shared molecular phenotypes. Discovering groups of patients whose disease shares molecular vulnerabilities, regardless of genotype or morphology, is the domain of statistical learning, and in this domain, it is an open research question how best to quantify uncertainty.
Ultimately, the randomized clinical trial provides the most powerful evidence in medicine, but trials are expensive, slow, and present unique ethical challenges, especially in children. Interpretable and rigorous statistical approaches are required; sick children will never be entrusted to “black box” machine learning approaches by their physicians.
We believe that cancers of the blood can serve as a model for more genomically and physically complex solid tumors, and we believe that insights from pediatric disease can impact both the understanding and treatment of adult disease. Children are not just small adults; the evolutionary history of pediatric disease is necessarily different from that in adults. The quiet genomic landscapes of pediatric cancer forces us to look closely, and in doing so, we can uncover subtle but fundamental forces shaping malignant evolution. For every malignant cell, a premalignant cell is among its ancestors, with a tiny yet crucial difference marking the change.
Cancers of the blood happen to be the best characterized in this respect, with clonal hematopoiesis (a loss of diversity in the millions of blood cells produced every minute) being an active topic of discussion: when does prevention become hypochondria? This, too, will eventually yield to analysis, with (one hopes) useful lessons for other types of cancer. Though our primary expertise is in hematological (blood) cancers, we also collaborate broadly with physicians to study a variety of pediatric solid tumors, to understand how liquid tumor biology can illuminate solid tumor treatment, and where fundamental differences prevent this.