Towards Interpretability and Reproducibility of Machine Learning Applications in Clinical Research