The aim is to identify methods for generation of causal hypotheses, using data mining techniques. Therefore, this will not involve testing specific hypotheses, but rather identifying new hypotheses from a large search space. This search space consists of an exposure subset and outcome subset, and the approaches will be flexible in the types of associations which can be found (e.g. non-linear). The exposure variables will be a set of Mendelian randomisation indicator variables - scores derived from genetic variants which have be previously identified as associated with a particular trait. We will use a large subset of available ALSPAC data as the outcome variable set. This project will include ensuring robustness against the issues of multiple hypothesis testing and also issues related to using a BMI score such as ensuring the association is not due to pleiotropy.