Prediction of OTX015 (MK-8625/Birabresib) drug response in cancer cell lines using machine learning model

Objective: Optimizing experimental processes, significantly reducing time and costs, and contributing to personalized treatments in healthcare, the focus has been on developing machine learning models to predict IC50 values representing the compound response. These models were applied to evaluate the performance on OTX015, which functions as a Bromodomain Extra Terminal Inhibitor, as well as on six different compounds (RVX-208, PFI-1, JQ1, I-BRD9, I-BET-762, and AZD5153). Method: Gene expression levels of cancer cell lines were used to create regression models using the R programming language to predict the IC50 values for OTX015 and other compounds concerning cancer cells. The models underwent enhancement through hyperparameter optimization. Trained models were tested using validation data they had not previously encountered. The models' performances were assessed using mean absolute errors, which represent the average of the squared differences between the actual and predicted values. Results: The study, through the analysis of machine learning models using the gene expression levels of cancer cell lines and compound IC50 values, demonstrates that particularly the SVM model exhibits consistent and generalizable performance with low error scores for both pre-laboratory studies on cancer cell lines and predicting personalized IC50 values. Conclusion: The results highlight the utility of machine learning models in drug discovery processes and personalized medicine. These models can provide rapid results and personalized IC50 values, thereby reducing costs. However, for more reliable outcomes, it may be necessary to further develop the models, gather additional data, evaluate them with different compounds, and conduct experiments with real-world data. Key words: machine learning, prediction methods, cancer, ic50, r programming