What methods can we use to compare networks (PPI, gene regulatory etc) created from single-cell gene expression (or proteomics) data?

There are methods that construct networks by comparing two conditions (e.g. from (gene) co-expression, or differential gene correlation (DGCA, Differential Gene Correlation Analysis)), and infer properties based on these networks.

What I'm interested here is the comparison of networks constructed for each single cell. The networks can be constructed solely from the data, or using known interactions (edges) with the help of a database. These comparisons could be used for understanding or classifying cells.

This will probably require a network construction method (which may be hard because we do not have replicates for single cells), some form of network similarity measure / network alignment method. This is a very hard task in general, but here we have 1-to-1 correspondence for each node between the cells (networks), which should help.

Similar discussions at

One related paper for single-cells is Chan et al., 2017, which uses groups of cells (created by other methods, e.g. clustering) to construct networks, and manually compares them (common edges / Venn diagram).

  • $\begingroup$ Could you clarify if you want a tool/methods to build those networks for each single cell or a tool/methods for comparing the cells according to the network you already have (ie assuming you already have a network for each cell)? if you ask about both it might be too broad, and that each question should be on its own to get good answers. $\endgroup$
    – llrs
    Commented May 29, 2018 at 15:13
  • $\begingroup$ Well, I'm interested in both. As there seem to be no such methods yet, I think we can leave the question as is and make a new one when and if this will be answered -- and need further discussions. $\endgroup$
    – Peter
    Commented May 29, 2018 at 15:19
  • $\begingroup$ Could you please clarify what type of comparison you mean. Are you only interested in common/different edges? Or are you also trying to compare network topology, shortest path analyses, centrality, network partitioning (what method?), etc etc. There are many metrics you can calculate for networks. If you just want to find the shared edges, it is trivial, assuming you have the networks. Is that really all you need? $\endgroup$
    – terdon
    Commented May 29, 2018 at 16:55
  • $\begingroup$ I cannot answer this: I believe what you are asking is the essence of my question; i.e. what metrics are available, and which of these "comparisons could be used for understanding or classifying cells". $\endgroup$
    – Peter
    Commented May 30, 2018 at 9:47

1 Answer 1


You can't build a network of a single cell only with the expression of a single cell. You either need previous known interactions or pathways or you need to use several cells/samples. If you use previous known information, you can use pathway information, otherwise you can group some cells and use something along the lines of WGCNA to find a scale-free network.

In the first case one can add the expression upon it and see what is happening, what edges are kept and which ones not...

Once you have the network for each cell you can compare the metrics for each cell. Some metrics to look at are the centrality of each gene and the betweneess.
In a more general way, you can look if it follows a scale-free topology or other type of network structures, or how many nodes are present, how many edges are present...

  • $\begingroup$ Thanks for clarifying that it's not possible to build a network from data of one cell. I'd prefer using some reference interactions database anyways -- why ignore that knowledge. $\endgroup$
    – Peter
    Commented May 30, 2018 at 9:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.