The ultimate goal in the field of personalized medicine is matching the best possible anticancer treatments to disease characteristics of individual patients and it is now clear that response to therapeutics depends not only on properties of cancer cells, but is also influenced by normal cells, molecules and tissues that surround and feed a tumor, so called tumor mircoenvironment. A new research recently published on Nature shows that artificial intelligence and machine learning can merge clinical, molecular and digital pathology data to predict response to breast cancer therapies, thus helping in identifying the best possible anticancer treatment for each patient: a relevant first step towards a translation of molecular, clinical and pathological information into a form that could be easily interpreted and utilized by clinicians.
For this study, professor Carlos Caldas from Cambridge University collected clinical, digital pathology, genomic and transcriptomic profiles of pre-treatment biopsies from 168 breast cancer patients treated with chemotherapy +/- HER2-targeted therapy prior to surgery. This information was subsequently linked to therapy outcomes (complete response or residual disease) at the time of surgery and used to develop a machine learning model which was applied to an unrelated cohort of 75 patients to verify its predictive accuracy.
Results show that this multi-omic approach based on AI is able to predict complete response with better accuracy than clinical features only: researchers found that malignant cell, immune activation and evasion features were associated with treatment response and that these features are derived from clinico-pathological variables, digital pathology, DNA and RNA sequencing. Using these data as input into a machine learning approach resulted in a multi-omic predictive model capable of analyzing multiple data; combining these molecular data with pathology information and other features that can be investigated before treatments provide unprecedented insights into the disease and help predict outcomes of different treatments in more detail and with better accuracy than even most experienced clinicians. «The framework highlights the importance of data integration in machine learning models; it could be also applicable to other cancers and could be customized to include both fewer and newer features», authors conclude.
