A machine learning algorithm may help to improve first-line treatment decisions in up to 17% of patients diagnosed with multiple myeloma (MM), according to a paper published in PLoS One.
MM is an incurable hematologic cancer with a mean estimated overall survival of 5 years. The disease is, furthermore, considered difficult to treat because of its heterogeneous clinical presentation, which may be related to chromosomal abnormalities.
It is, however, unknown which chromosomal abnormalities in MM are linked with greater treatment sensitivity. This is further complicated by the number of treatment options available in the first line of therapy. A goal of MM research, therefore, is to link biological and genomic markers with treatment sensitivity, which would in turn help to improve outcomes in this patient population.
Machine learning has helped to improve several areas in medical care, including diagnosis and some aspects of treatment in oncology. For this study, researchers developed a Multi Learning Training approach (MuLT) algorithm—a tool which evaluates combined clinical and genomic data—to determine whether machine learning could help to better direct treatment recommendations among patients with MM.
Data from 1525 patients with newly diagnosed disease were included. All data were obtained from the Multiple Myeloma Research Foundation (MMRF) CoMMpass study.
Analysis of use of the MuLT algorithm suggested that genomic profiling can recapture molecular damage noted in fluorescence in situ hybridization analysis. Cross-validation experiments showed, furthermore, that the algorithm could predict MM treatment sensitivity effectively, with an area under the curve of 68.7%.
The authors noted, finally, that MuLT could have predicted superior first-line treatment to that received in 17.07% of evaluated patients.
“Next steps are related to applying MuLT over different cancer data sets composed of clinical markers, gene expression levels, and treatment,” the authors wrote. “This study was limited to binary classification, stratifying patients into either treatment sensitive or non-sensitive, but it could be generalized to perform regressions and multi-categorical classification.”
Venezian Povoa L, Ribeiro CHC, Silva ITD. Machine learning predicts treatment sensitivity in multiple myeloma based on molecular and clinical information coupled with drug response. PLoS One. 2021;16(7):e0254596. doi:10.1371/journal.pone.0254596