Scientists in Singapore, from Duke-NUS Medical School, A*STAR’s Genome Institute of Singapore (GIS), and Singapore General Hospital (SGH), alongside their colleagues, have harnessed the power of artificial intelligence (AI) to expedite the identification of crucial biomarkers that can predict the non-responsiveness of chronic myeloid leukemia (CML) patients to conventional treatments. This early prognosis enables patients to receive potentially life-saving bone marrow transplants during the initial stages of the disease.

CML is a form of blood cancer triggered by a genetic mutation that permanently activates an enzyme called tyrosine kinase. This mutation occurs in a blood stem cell within the bone marrow, causing it to transform into an aggressive leukemic cell that eventually supplants healthy blood production.

The standard treatment for CML involves tyrosine kinase inhibitors (TKIs), which deactivate the overactive enzyme resulting from the genetic mutation. However, individual responses to these drugs vary significantly. Some patients respond exceptionally well, leading to life expectancies comparable to normal individuals. On the other end of the spectrum, certain patients exhibit minimal response, and their disease progresses to an aggressive state known as blast crisis, which is resistant to all standard therapies. Since bone marrow transplantation is the sole treatment option for blast crisis, performing it during the early stages of the disease yields the highest effectiveness. Detecting TKI resistance in patients sooner can mean the difference between survival and premature death.

Associate Professor Ong Sin Tiong, the study’s senior author from Duke-NUS’ Cancer & Stem Cell Biology Programme, emphasized, “Our work indicates that it will be possible to detect patients destined to undergo blast crisis when they first see their hematologist. This may save lives since bone marrow transplants for these patients are most effective during the early stages of CML.” Dr. Vaidehi Krishnan, Principal Research Scientist with the CSCB Programme and first author of the study, added, “Based on these findings, we aim to develop simple clinical tests that can advise physicians on the optimal choice of treatment at the time of diagnosis.”

The research team leveraged single-cell analysis coupled with the power of AI to predict drug response in leukemia. By generating a cell ‘atlas’ from bone marrow samples taken from healthy individuals and CML patients prior to treatment, the team obtained insights into the different cell types and their proportions in each sample. They performed single-cell RNA sequencing and employed machine learning algorithms to identify the genes and molecular processes activated or deactivated in each cell.

The study revealed eight statistically significant features in pre-treatment bone marrow cells that were associated with either sensitivity or extreme resistance to tyrosine kinase inhibitor treatment. Patients showed a higher likelihood of responding well to treatment if their bone marrow samples exhibited a stronger inclination towards premature red blood cells and a specific type of tumor-destroying ‘natural killer cell.’ As the proportions of these cells in the bone marrow changed, patient responses to treatment varied.

This research could pave the way for identifying drug targets to prevent or delay treatment resistance and blast crisis in patients with chronic myeloid leukemia. Associate Professor Charles Chuah, from Duke-NUS’ CSCB Programme and Senior Consultant at the Department of Hematology, SGH, and NCCS, expressed optimism about the improved treatment outcomes for CML patients, stating, “Knowing which treatment works best for our patients will further enhance these outcomes, and we are excited by the possibility of being able to do so.”

The next step for the team involves developing a predictive test for treatment resistance based on these findings, which can be routinely employed by hospitals in their patient care protocols.

By Impact Lab