Which therapy is best for which patient?
Personalized antineoplastic treatment is a crucial advancement in cancer management tailoring therapies according to individual patient biology. However, most commonly patients are still treated with a ‚one-size-fits-all‘ chemotherapy regimen since deciphering tumor biology and identifying potential markers of adverse risk is a challenging task and targetable molecular lesions for whom specific drugs exist are rare.
In our lab, we use supervised machine learning to predict response to therapy given individual patient biology. Our models are trained on large multi-center data sets of hematologic malignancies such as acute myeloid leukemia, where we currently cooperate with the German Study Alliance Leukemia, the German-Austrian AML Cooperative Group, and the European HARMONY consortium.
Our models automatically identify important patient features for outcome prediction, such as molecular and cytogenetic alterations, to provide a data-driven basis for patient management.
As of now, our AML model predicts achievement of complete remission with an accuracy of 86%. Additionally, our model has identified a novel marker of adverse risk in AML that has previously been unacknowledged (but we do not want to spoil the suspense until final publication).
We now aim to expand our models to other hematologic and oncologic neoplasms such as acute lymphoblastic leukemia, multiple myeloma, diffuse large b-cell lymphoma, lung cancer as well as colorectal cancer (among others). Stay tuned for more exciting advancements in machine-learning-guided response prediction!
Eckardt JN, Röllig C, Metzeler K, et al. Prediction of complete remission and survival in acute myeloid leukemia using supervised machine learning. Haematologica. 2023;108(3):690-704. Published 2023 Mar 1. doi:10.3324/haematol.2021.280027