 Hi everyone, today I'm going to be talking about net coupler, which is an algorithm in our package for inferring causal pathways between high dimensional metabolomics data and external factors. Our motivation for designing and creating net coupler was to be able to make use of moderately high dimensional and complex network data, and to be able to answer questions about causal pathways that involve that network. As illustrated by this figure or diagram. We want to know how an exposure like exercise might influence the network, how the network might influence an outcome like diabetes, or how an exposure might influence an outcome through metabolic network. There are several main features of net coupler in that it can find that likely network structure underlying metabolic data, you can include exposure and outcome data, and it finds that those potential causal links that connect the exposure outcome and the network. It's also quite flexible in the type of models that you can use within it so you can use linear regression models or Cox proportional hazard models. The models are used, you can also adjust for potential confounding factors that might bias your results. And that coupler is designed as to work with network graphs so it's the results are designed to be also visualized like that so for instance through tidy graph or G graph packages. There are four basic phases of the network algorithm. The first phase is to derive the underlying metabolic network structure. And the second phase is then to use that, that metabolic network structure and iteratively select the metabolic variables within called nodes, and sets each node as an index node and select the neighboring nodes of that index node. And the third phase that would be three neighbors for this index node. The third phase is to then calculate all possible combinations of index node with the neighboring nodes. So, and then set the and then use them within in the model. In this case there are three neighbor, three neighboring nodes, that would be eight different combinations. The fourth phase is then taking those models and linking them with the exposure. E, or the outcome, oh, and then based on some specific threshold, classifying the link between exposure or outcome and the index node as either direct effect ambiguous or no effect. So the physical model output can allow for visually inferring about potential causal pathways. So for instance, a net coupler might identify to direct effects here that that are the thicker lines of the arrows, or to end to ambiguous effects which are the thinner lines from the exposure or towards the outcome. And then you can visually trace the paths that go from the exposure to the outcome through the network graph, and you can identify specific metabolic variables here marked as red as potentially being on the causal pathway between the exposure and the outcome. We're currently actively working on this on this package so there are some sets of limitations and areas to improve. They're quite tricky to visualize when you start to have too many variables so that's something that we're working on improving. Because the data has been pre-processed beforehand, the model output estimates are difficult to interpret. We also don't believe that this algorithm is best suited for purely explorative purposes. There should be some theoretical basis to the research question. The modeling also relies that the classification threshold also relies heavily on the P values so we're working on using different types of thresholds. And we've only tested this on cross sectional and time to event data so we don't know how it works in other settings. It's low so we haven't tested on networks with larger than 25 variables and it's probably not suitable for very high dimensional data. Anyway, if you're interested in learning more, the package link is in the footer. Thanks for listening.