Biological markers or “biomarkers” have profoundly influenced medical research and clinical care. Depending on their association with specific disease pathophysiology, biomarkers can serve to determine the risk of developing a disease, screen for subclinical disease and diagnose overt disease. Such biomarkers may also identify drug targets, if the biomarker is causally associated with the disease.Current approaches employed for the discovery and validation of biochemical biomarkers of disease involve the design of expensive and time-consuming studies that use precious tissue samples from humans and animal models. Further, explorative biomarker research generally employs only classic observational designs such as case-control and cohort studies, which can suffer from bias introduced by confounding and/or reverse causation, hindering a causal interpretation when exploring etiologic biomarkers. The identification of etiologically-involved biochemical biomarkers is therefore lengthy, costly, technically challenging and, thus slow overall, ultimately limiting the number of biomarkers that translate into clinical care. Novel strategies are hence needed upstream of these costly steps for two main reasons: (1) a more effective identification of candidate biomarkers of disease and (2) a more robust evaluation of candidate biomarkers via complementary and independent lines of evidence that are subject to different key sources of bias. Metabolites—such as sugars, lipids and amino acids—are a large class of biochemical biomarkers that can be measured en masse through emerging metabolomics platforms and could thus help to identify new functional biomarkers. In turn, since the genetic determinants of several metabolites have been identified, this information can be applied to understand the role of metabolites in disease causation and to help inform the design of classic observational studies, which in combination could offer a robust approach to evaluate them. In this doctoral thesis, I explored a class of candidate small molecule blood biomarkers called metabolites, which are intrinsically involved in the human organism’s normal function and, thus their change can reflect disease pathophysiology. Given that metabolite distribution and availability is dictated by enzymes and other proteins coded in the human genome, blood metabolite levels—like many diseases—have a genetically heritable component that has been studied and reported through genome-wide association studies (GWAS). Here, I offered candidate etiologic biomarkers of three diseases—breast cancer, idiopathic pulmonary fibrosis and lung cancer risk—through a novel approach that started by identifying shared genetic determinants of both blood metabolite levels and disease risk using available GWAS summary statistics for both traits. Then the genetic variants jointly associated with metabolite blood levels and disease risk were used as instruments in a type of instrumental variable analysis called Mendelian randomization (MR) that uses genetic variants as instruments using, in this case, the genetically determined component of candidate metabolites to test and estimate their causal effect on disease risk, while—in principle—avoiding confounding by life course factors as experience by classic observational associations.In contrast to the putative application of MR to test hypotheses sourced from classic observational and other designs, the approach presented in this thesis used MR as a hypothesis-generating tool. Metabolites identified via genetics and MR later served to inform the design of separate casecontrol studies to estimate the effect of the identified candidate metabolites from an independent approach in line with the tenets of triangulation.