Author:
Yin Liangying,Liu Menghui,Shi Yujia,Qiu Jinghong,So Hon-cheong
Abstract
AbstractAccurate identification of direct causal(parental) variables for a target is of primary interest in many applications, especially in biomedicine. It could promote our understanding of the underlying pathophysiological mechanism and facilitate the discovery of new biomarkers and therapeutic targets for studied clinical outcomes. However, many researchers are inclined to resort to association-based machine learning methods to identify outcome-associated variables. And many of the identified variables may prove to be irrelevant. On the other hand, there is a lack of an efficient method for reliable parental set identification, especially in high-dimensional settings (e.g., biomedicine).Here, we proposed a novel and efficient two-stage approach (I-GCM) to discover the direct causal variables (including genetic and clinical variables) for various outcomes. Variable selection was first performed by the PC-simple algorithm. Then it exploited the invariance of causal relations in different (experimental) settings, which was represented by generalized covariance measure calculated from gradient-boosted trees, for efficient and reliable causal variable discovery.We first verified the proposed method through extensive simulations. This approach constantly yielded high precision (a.k.a., positive predictive value) and specificity while maintaining satisfactory sensitivity in general, and consistently outperformed a standard Notably, the precision was larger than 90% in our simulated scenarios, even in high-dimensional settings. We then applied the proposed method to 4 clinical traits to uncover the corresponding direct causal variables. Encouragingly, many identified clinical variables, genes and pathways were supported by the literature. Our proposed method constantly achieved superior performance in identifying actual direct causal variables, making it particularly useful in selecting what (genetic/clinical) risk factors to follow up. Importantly, our work represents one of the first applications of the invariance principle for causal inference in biomedical or clinical studies, and suggests a new avenue for causal discovery in these settings.
Publisher
Cold Spring Harbor Laboratory