Future Direction

One of our future directions is to make visualizations of the predicted ligand-receptor pairs and interactions from NicheNet. A great visualization tool to be explored is circos plot, which displays the data in a circular layout so that it has advantages of exploring interactions between objects or positions on the circle clearly and presentably.

Another direction to consider is to incorporate single human microglia cells from multiple batches, eliminate the batch effects, leave greater amounts of pure cells of a single cell type with less noise to locate potential ligand -receptor interactions from the background noise. If the ligands inputted can produce a significant amount of signals with limited noise, the predicted ligand-repector pairs would be more confident.

Besides, we might consider making the pipeline more automated with machine learning methods, so that researchers can find predicted receptor ligand interactions of high-confidence without self-defining and testing unique parameters or settings for each experimental design.

Integrating Data from Multiple Sources

Create our own graph based, machine learning pipeline for signal transduction networks rather than building tools off of existing ones. Current algorithms seldom use modern machine learning methods and focus more on regressions and other simple statistical measures applied to graphs. Different approaches exist but generally, genes are assigned to nodes and interactions are assigned to weighted edges. Packages for optimized graph traversal algorithms are commonplace and could be leveraged to create an accurate and efficient tool. If large datasets are available, a neural network could be very effective. If not, other machine learning statistical inference methods could be utilized.

Future Machine Learning Concept