Multiple Myeloma is an incurable and deadly hematological cancer. Globally, its impact is esti- mated around 180 000 deaths annually. One of the reasons for this mortality is a myeloma related complication called bone disease. Bone disease is manifested in around 80% of multiple myeloma patients.

In this thesis, we developed a biomarker for stratification of patients at risk of bone disease in multiple myeloma. For this purpose, we integrated a web data mining method called NetRank, with the machine learning algorithms Support Vector Machines and Random Forest, and then applied them to a cohort of 85 myeloma patients.

My results was a biosignature for stratification of at risk bone disease patients with accuracy of 85% and F1-score of 88%, showing that the model could separate the two classes and make reliable classifications in both. This biosignature only consisted of 25 features, 22 transcriptomic features and 3 clinical features, showing its potential for clinical applications. To ensure that the models results are not due to the model overfitting, both the internal crossvalidation results and the external test- set results were analysed. In short, a biosignature for risk stratification of bone disease patients in multiple myeloma is discovered that could be applied in clinical settings for real-world validation.