On the (un)reliability of learning Gaussian graphical models from microbial abundance data
I explored the reliability of microbial network inference. This topic, which is gaining popularity, has many overlooked issues and open challenges. In particular, I dug into the usage of Gaussian graphical models. I evaluated the performance of SPIEC-EASI (a popular method with more than 1300 citations at the time of writing) and a novel Bayesian method based on BDgraph. I concluded that most analysis would require in the order of thousands of samples (which is often not the case in environmental microbiology studies).
The report, slides and all source code are publicly available on GitHub if you want to know more!