Yes, click-blait title, but I found the contents somewhat interesting. There was a science article on these guys' same algorithm and they pointed out that a lot of modeling efforts rely on either linear or high-order, but heavily constrained equations.

This breaks down in most real systems where even small amounts of non-linear relations between variables causes erroneous predictions of cause and effect (i.e. correlation does not imply causation)

What's interesting is their approach turned traditional definitions of cause and effect on their head. One version of this goes that X "Granger causes" Y if the predictability of Y goes down when you remove X from your equations. However CCM, the algorithm pictured at the top of the article, says that X causes Y if you can estimate X using the historical records of Y.

It's a pretty simple switch, but it means that if you have long enough time-series information about a system, you have a pretty general way to infer relationships in it. My friend who invited me to the discussion wants to try using it to tease apart relationships of bacteria in the gut, where you can collect many poops across a person's trial, but have almost no direct tools to observe or manipulate bacterial interactions.

Sadly time series data is a bit harder to come by for sub-cellular protein interactions, otherwise I'd be ready to try it out on my own experiments =/