Add PHONEMeS-ILP

[agitter]

PHONEMeS: Efficient Modeling of Signaling Networks Derived from Large-Scale Mass Spectrometry Data https://doi.org/10.1021/acs.jproteome.0c00958

https://github.com/saezlab/PHONEMeS-ILP

From the readme

PHONEMeS requires the interactive version of IBM Cplex or CBC-COIN solver as the network optimiser.

[ctrlaltaf]

@agitter @annaritz I've been reading up on this method and I have a few questions/points regarding the implementation of phonemes to spras which hopefully is a good start to researching this method and CPLEX compatibility.

• The current version of the PHONEMeS-ILP is an R package, which was last updated two years ago. Since spras is currently mostly written in Python, is integrating R into spras a possible research point, or is there a way to make this R package compatible with the current Python spras environment?
• PHONEMeS seem to have three main function that support different ways mass spectrometry data (runPHONEMeS, runPHONEMeS_mult, and runPHONEMeS_UD) which incorporates varying data inputs. I still don't fully understand the way that it takes in inputs, but maybe this is something to discuss related to the current compatibility of spras.
• One drawback listed from the paper is the requirement of prior knowledge networks as an input, and maybe this is something to consider when trying to implement phonemes

[agitter]

A general comment for PHONEMeS and other methods is that these issues are an initial suggestion about pathway reconstruction algorithms we may want to include in SPRAS. After studying the method, we may conclude that their use case is too specialized or that SPRAS is not general enough to support their input data currently and that they then cannot be added.

Our approach to wrapping algorithms puts them all in a Docker container so that we don't have to consider interactions between our Python code and whatever language the pathway reconstruction algorithm in implemented in. Our Docker image would have an R environment to run PHONEMeS. The problem we find with some R packages is that they assume interactive analysis and do not provide a command line interface. That means we would have to write a small R script that is callable from the command line and include that in the Docker image too.

I would need to re-read the paper to remember what different data inputs each of these three modes requires. If we find that one or more of them is not compatible with SPRAS (e.g. the multi-time point mode), we could initially one support those that are. Lisa's initial research into KeyPathwayMiner #110 indicates that it also expects a matrix of node scores for nodes X conditions as opposed to a vector of node scores.

The prior knowledge network is what SPRAS considers to be the background network. From their supplementary text:

In PHONEMeS, the interactions are considered to be directed (kinase-to-substrate) and unsigned.

So it appears to be an unsigned, directed graph, which we will be able to support after #107. However, we need to read more carefully to see whether the edges need to be annotated with substrate information that specifies the phosphorylation site. That would make this method more specialized that others in SPRAS.