Machine Learning for Predicting Therapeutic Outcomes in Acute Myeloid Leukemia Patients

• DOI: 10.1101/2024.02.29.24303536
• Publication Date: 2024-03-03T00:20:20Z
• Authors: Nestoras Karathan...; Panayiota Papasavva; Anastasis Oulas; George M Spyrou
• Keywords: Exome; Predictive power

Abstract Background and Objective The standard of care in Acute Myeloid Leukemia patients has remained essentially unchanged for nearly 40 years. Due to the complicated mutational patterns within between individual a lack targeted agents most events, implementing individualized treatment AML proven difficult. We reanalysed BeatAML dataset employing Machine Learning algorithms . project entails extensively characterized at molecular clinical levels linked drug sensitivity outputs. Our approach capitalizes on data provided by predict ex vivo 122 drugs evaluated project. Methods utilized ElasticNet, which produces fully interpretable models, combination with two-step training protocol that allowed us narrow down computations. automated genes’ filtering step two metrics, we all possible combinations identify best configuration settings per drug. Results report Pearson correlation across 0.36 when RNA sequencing were combined, best-performing models reaching 0.67. When trained using datasets isolation, noted Sequencing (Pearson: 0.36) attained three times predictive power whole exome 0.11), falling somewhere (Pearson 0.26). Lastly, present paradigm significance. used our models’ prediction as health management score rank an individual’s expected response treatment. identified 78 out 89 (88%) proposed was more potent than administered one based their data. Conclusions In conclusion, reanalysis demonstrates potential patients, addressing longstanding challenge personalization this disease. By leveraging data, yields promising correlations predicted actual responses, highlighting significant forward improving therapeutic outcomes patients. Highlights learning can are informative predicting Drug predictions could be 88% (78 89) examined.