On the relationship of machine learning with causal inference

Excerpt

There is rapid emergence of increasingly sophisticated informatics being applied in biomedical research. Machine learning is a collection of algorithms (including deep neural network learning, support vector machine, random forest, and super learning which integrates all above algorithms) often used in informatics and has been successfully applied in clinical decision making, diagnosis, and image identification [1, 2]. Given its success in these areas, there is great anticipation that a similarly impressive impact is to be expected in explanatory research, the area of research focused on identifying and understanding causes of disease. However, application of machine learning in explanatory research is not straightforward and might lead to misinterpretations [3]. Efforts are ongoing to develop an appropriate theoretical context for integrating machine learning with explanatory research [4]. In this Letter, we outline the fundamental differences between predictive research and explanatory research and summarize challenges and possibilities of applying machine learning in explanatory research. …