Data-driven Approaches to Explore Precision Medicine

Precision medicine in a contemporary context implies customising healthcare based on individual biomarkers, such as genetic variants or lifestyle factors. The purpose is to prevent, diagnose or find the most effective disease treatment approaches customised for the individual or subgroups of patients instead of a one-size-fits-all approach.

As a systemised approach, this concept has come into focus in recent years in modern translational science; however, it should be noted that ancient medicine systems such as Ayurveda, a traditional medicine in India, has over centuries of history looking into patient stratification in relation to disease development and treatment, and a fairly layered system to describe it that incorporates elements of lifestyle, behaviour, diet and proxy biomarkers for underlying genetics.

In this PhD thesis, I have 1) explored precision medicine concepts from different perspectives; 2) used different approaches to analyse patient clinical (application note, chapter 8) and genomics data; 3) utilised genetics from genome-wide association studies and next-generation sequencing analysis; 4) developed stratification-based models, such as Ayurveda-based deep phenotyping, polygenic risk scores, and machine learning models; and 5) discussed how these models could be applied in a clinical setting for prediction of phenotypes, treatment response and late-side effects.

The first paper, presented in chapter 4, explores the use of Ayurveda medicine for patient stratification to help identify novel disease genetic variants that predispose towards rheumatoid arthritis. The second paper, chapter 5, uses two developed and validated adult cancer polygenic risk scores to explore risk stratification for different phenotypes in childhood cancer.

The third and fourth papers, chapters 6 and 7, respectively, focus on the development of machine learning models to predict treatment late-side effects, specifically, cisplatin-induced hearing loss and nephrotoxicity, respectively, in testicular cancer patients, using clinical and genomics data. In chapter 9, it is presented a model that predicts dasatinib treatment response in T-cell acute lymphoblastic leukaemia.

This work was developed at St. Jude Children’s Research hospital during my external stay. These stratification-based models may help leverage heterogeneous clinical data and find disease-associated genomic markers. Furthermore, implementing these models in a clinical context, together with medical expertise, may allow for earlier disease diagnosis, personalised prevention, and treatment strategies for groups of people based on their genomics and clinical profiles.

Ultimately, this will enable a better balance between treatment efficacy and patient’s quality of life.