Hello all, I sent this to Paramvir and Ben, but I realize it may be of general interest. It's just an enhanced detailed workflow of the phenotype reconciliation process I think we need to really handle Adam D.'s data. Chris Begin forwarded message:
From: Christopher Henry <[email protected]> Date: October 10, 2012 12:43:03 AM CDT To: Ben Bowen <[email protected]> Cc: Paramvir Dehal <[email protected]>, Ross Overbeek <[email protected]> Subject: Phenotype reconciliation workflow
So I've been thinking this over. I don't think the problem is so difficult or underdetermined as we originally thought. Check out the attached workflow (start at purple box... red and green boxes are endpoints). This is what we need to implement to handle Adam D.'s data, and while it looks complex, I actually don't think it's that bad... each individual step is pretty straightforward, and many steps don't involve incredibly complex modeling (I particularly like the idea of seeking reaction essentiality patterns that match gene essentiality patterns...). I think our mistake was to think we should try and handle all phenotypes the same way... We have three distinct classes of phenotypes that should be handled differently:
1.) Wildtype growth condition phenotypes - the Biolog scenario (pretty easy to handle) 2.) Gene essentiality phenotypes (harder to handle, but offers chance to make new annotations) 3.) Multi-gene KO phenotypes (very tough to handle... almost certainly must be handled one-by-one manually... at least until we have good pipelines for (1) and (2)
So we just handle things in the right order:
1.) Wildtype phenotypes 2.) Single gene KO phenotypes -Genes with confident annotations -Genes with unconfident or no annotations 3.) Multigene KO phenotypes
On Oct 9, 2012, at 8:48 AM, Ben Bowen wrote:
Agreed. Just a little bit of book-keeping will be of huge service -------------- Email: [email protected] Phone: (510) 214-6683 MS 977-221, Lawrence Berkeley Lab Web: ms.lbl.gov
On Mon, Oct 8, 2012 at 10:47 PM, Christopher Henry <[email protected]> wrote: I agree the mapping of genes based on phenotypes is a concern. I think it should be possible, but we'll need to combine the modeling with some heuristics based on pre-analysis of the data. Chris
On Oct 8, 2012, at 5:26 PM, Ben Bowen wrote:
I'm tingling with anticipation. FYI: I gapfilled to grown on alanine as C and N source and it added alanine ABC transporter. seems reasonable.
There is a layer that we need to put something together for. Given a list of gap-filled reactions, go back to the gene-fitness data and identify candidate genes that are responsible for the function. If you look at the matrix of fitness data and some added reactions, can you really pick out which gene is responsible? In the alanine example above, there are 100s of essential genes, and many won't have known function. I've been trying to sort out the mechanics of this. Hopefully, on Wednesday, I'll have some kind of rational approach sorted out.
-------------- Email: [email protected] Phone: (510) 214-6683 MS 977-221, Lawrence Berkeley Lab Web: ms.lbl.gov
On Mon, Oct 8, 2012 at 2:58 PM, Christopher Henry <[email protected]> wrote: Sorry I missed this call today. I've had an extremely busy couple of weeks. I've been working alot on the gapfilling commands, with some big improvements that you should appreciate Ben. I'm working on writing detailed tutorials of the new process, but essentially gapfilling is now split over two steps: (i) gapfill which identifies one or more solutions to resolve a false negative, and (ii) managesol which shows all current gapfilling solutions for a model and allows you to delete solutions or specify solutions for integration in the model. See the example output below.
Now you can begin to imagine the next step in implementing our reconciliation pipeline: batch gapfilling creates solutions for all false negative conditions, and managesol allows you to see and select solutions for implementation. I'll also add an ability to simulate all gapfilling solutions against phenotype data so they can be assessed based on their ability to replicate experiments. The final step will be to automate the selection of an ideal cross section of solutions to fit phenotype data.
1.) model gapfill testmodel --media Media2 ********************************************* Gapfilling formulation: GF9 Media: Media2 --------------------------------------------- 1 gapfilling solution(s) found. --------------------------------------------- New gapfilling solution: GF9.0 Solution cost:2 Gapfilled reactions 2{ ID Direction Equation rxn8[c] < (1) A name[c] <=> (1) D name[c] rxn4[c] > (1) D name[e] <=> (1) D name[c] } ---------------------------------------------
--Note: the solution is saved in the model automatically, but not yet integrated.... you view and integrate solutions using the "managesol" command:
2.) model managesol testmodel ********************************************* No unintegrated gap generation studies! ********************************************* ********************************************* 10 unintegrated gapfilling studies! ********************************************* Gapfilling formulation: GF0 Media: Complete --------------------------------------------- 1 gapfilling solution(s) found. --------------------------------------------- New gapfilling solution: GF0.0 Solution cost:2 Gapfilled reactions 2{ ID Direction Equation rxn2[c] > (1) B name[e] <=> (1) B name[c] rxn8[c] > (1) A name[c] <=> (1) D name[c] } --------------------------------------------- ********************************************* Gapfilling formulation: GF1 Media: Media2 --------------------------------------------- 1 gapfilling solution(s) found. --------------------------------------------- New gapfilling solution: GF1.0 Solution cost:2 Gapfilled reactions 2{ ID Direction Equation rxn8[c] < (1) A name[c] <=> (1) D name[c] rxn4[c] > (1) D name[e] <=> (1) D name[c] } ---------------------------------------------
On Oct 8, 2012, at 12:58 PM, Ben Bowen wrote:
I would be happy to give a brief update on the phenotype simulations -------------- Email: [email protected] Phone: (510) 214-6683 MS 977-221, Lawrence Berkeley Lab Web: ms.lbl.gov
On Mon, Oct 8, 2012 at 10:53 AM, Paramvir Dehal <[email protected]> wrote: Hey guys,
Sorry for the short notice, we have a microbes call in 10min. The only thing I have for the agenda is to talk about keeping track of our needs and making sure we get help from Tom, Nomi and others. Let me know if you have additional agenda items.
Paramvir _______________________________________________ Microbe mailing list [email protected] https://lists.kbase.us/mailman/listinfo/microbe
_______________________________________________ Microbe mailing list [email protected] https://lists.kbase.us/mailman/listinfo/microbe