how to "link" two datasets?

[ergonyc]

The group discussion (Jan, Brain Trust Meeting) let do a consensus to refocus on extending the functionality of the browser tool IMMEDIATELY with the Ability to link across datasets.

How are we going to do this?

Caveat: we will look to "link" datasets only... though we should anticipate that on the roadmap (v3.0) we will want to harmonize and link / integrate across -omic domains.

• What is the linking proposition?
  Q1: create an NDCN "linking table"/algorithm vs. a preferred API?
  • potential APIs?
  • omic-specific tables?

Q2: Potential Experts on "linking" or -omic ontologies to leverage?

[ergonyc]

@ddifraia @ricoderks @bnovotny any thoughts?

[ergonyc]

@MenonLab shared this perspective via e-mail:

Modality 1	Modality 2	Basic	Advanced
Proteomics	Transcriptomics	Gene-to-protein name mapping	Isoform-level mapping of genes and proteins
Proteomics/Transcriptomics	Lipidomics	Genes/proteins known to be involved in synthesis of individual lipids – mapped to each lipid	Proteins involved in general lipid regulation pathways – lipids involved in same pathways, aggregated by pathway
Proteomics/Transcriptomics	Metabolomics	Proteins known to be involved in synthesis/regulation of individual metabolites – mapped to each metabolite	Proteins involved in general metabolic pathways – metabolites involved in same pathways, aggregated by pathway
Lipidomics	Metabolomics	Lipids mapped to metabolic products and vice versa (non-unique mapping)	Lipids and metabolites involved in same general pathways, aggregated for each pathway

[ricoderks]

I think during the curation we already need some form of general ID number or the proteins/genes/lipids/metabolites so that we can use them.

[ddifraia]

@ergonyc I think the best way in this case would be to download species specific tables. For human, mouse, and at least some of the key model organism. Without having to connect to the database everytime. In this case the user would need to specify the species of each experiment thou...

[MenonLab]

Lipidmaps looks very interesting indeed, and would be a nice way to map between lipid species and genes/proteins. I've been looking around to see if there is a static table we could download from there

[ricoderks]

@MenonLab you can download it. It's a txt file of about 3.5-4Mb. The only thing is, I just saw, that the latest timestap is from 2015!

@ddifraia @ergonyc I agree, the best thing is to download them and update the tables a few times a year if nescessary.

[MenonLab]

@ricoderks - that's great! Even if it hasn't been updated in a few years (2015 seems like a lifetime ago, haha), we can use it as a framework for testing ways to map.

[bnovotny]

Hi all, wanted to throw out a resource for linking metabolomics to proteomics and transcriptomics: the Comparative Toxicogenomics Database. Their interactions can be downloaded (https://ctdbase.org/downloads/). It would not be totally comprehensive but might be a good supplementary resource.

[ergonyc]

Cool resource. Do you think that we would want to have a master table to join against, or query the online DB on the fly?

[ergonyc]

What other resources are out there that ought to be covered?

[ricoderks]

I think the best is, if possible, to download. That avoids causing problems on our side if they change the api or whatever.

[ergonyc]

I agree with @ricoderks that for our first iteration of the linking we should start with a fixed and agreed upon table to join against. Who has two databases to develop against? Someone needs to code up a first attempt at "linking".

[ricoderks]

Is there a data set available?

[ergonyc]

Great insight today's "brain trust" discussion about two types of linking:

1. categorical / top-down linking
2. statistical (observed/inferred) linking

Are we in consensus that we should start with type 1 for now?

[ricoderks]

Agree!

[ddifraia]

I also agree

[ergonyc]

Has anyone tried a proof-of-concept to make such a linking?
@ricoderks I will send you a private link to some AnswerALS data which has transcriptomics AND proteomics.

@ddifraia @mtiessen1175 I'm assuming that you have some available proteomic/transcriptomic data sets you can test with?

@bnovotny ?

[bnovotny]

Hi, I've found a more comprehensive file for linking metabolites and proteins (from STITCH) so I can try a test of linking some of our metabolomic and proteomic data.

Have we pinpointed what "link" means at this point? Once we have a joining table, are we joining metadata? DE results? Count/abundance data? All of the above? Trying to visualize the final product of this "version 1" linking to determine how to proceed with a proof-of-concept.

[ricoderks]

@ergonyc I started yesterday to code something, but need a dataset.

[ricoderks]

@bnovotny Good question! I'll talk with Martin and Yassene and see what they think!

[ddifraia]

@ergonyc We have some transcriptome and proteome data, I'll need to ask Alessandro what is the best dataset we have in this regard. Probably some mouse data...