Causal inference for computational biology

In recent years, advances in technology and alternatives to treat patients increased interest in therapy recommendations that consider a patient's environment, lifestyle, and individual genes. In such a context, Causal inference methods can be a powerful tool in precision medicine studies. There is, however, a gap between biomedical applications and causality methods: the large number of covariates, unobserved confounders, small sample size, and multiple treatments evaluated simultaneously are common characteristics in biomedical studies and often a limitation in causality. In this thesis, I present three works. The first work proposes a method to estimate causal effects on applications with multiple treatments with a multi-task learning neural network architecture. The second work proposes a treatment effect estimator for small high-dimensional datasets that incorporate transfer learning techniques to improve the covariate adjustment. The last work proposes a method that incorporates partially known information to causality outputs to support the validation of methods on real world applications. Together, these works contributes to decrease the gap between causal inference and biomedical applications.