I am working with protein blocking assays and the end result is a 2D matrix describing which antibodies block the binding of other antibodies to the target antigen.I need to group the antibodies together into "bins" based on their combined profiles in both the row and column direction.I am able to group the blocking profiles of rows vs rows, or columns vs columns, using clustering. The end results could look something like this:>col_bins binAb1 1Ab2 2Ab3 2Ab4 2Ab5 3Ab6 4Ab7 5Ab8 5Ab9 6In this case the "bin" values are just to describe they have similar blocking profiles - so Ab2, Ab3, Ab4 have the same blocking profile, as do Ab7 and Ab8. Looking at the row profiles>row_bins binAb1 1Ab2 2Ab3 3Ab4 3Ab5 4Ab6 5Ab7 5 Ab8 6Ab10 7The important end result, where I am stuck, is how to combine this with the row direction and only report those that are represented in both directions AND group together in both directions. It is possible that some Abs will not be represented in both directions. The "bin" values of row_bins and col_bins are also not important, just the relationship between Abs by name that belong in the same bin, in both directions. In other words, a combined bins report would look something like this: binAb1 A Ab3 BAb4 BAb5 C I made this visually because it is clear that these are the only groupings that are maintained in both directions. But real data sets are much bigger, so I need some form of automation. Any ideas on how do this with matrix, dendograms or clustering functions? Thank you, josh [[alternative HTML version deleted]]
Josh, A couple of things: 1) It would be helpful if you can provide some reproducible data and the code you have developed thus far. 2) This is more of a stackexchange.com or crossvalidated.com question. That said...without seeing the data... Dendrograms/hclust are generated by using a distance matrix. In ecology distance would be measured by an index summarizing site similarity based on species composition. Selecting the appropriate index is crucial. Legendre and Legendre have a chapter on this in Numerical Ecology. You don't have species or sites, but you do have antibodies, antigens and blocked antibodies. You could treat antibodies as "sites" and how they block other antibodies as "species." So for your data, you might have two antibodies that block antibodies x,y, and z, but A1 blocks with a 0.2, 0.5, and 0.8 efficiency while A2 blocks with a 0.8,0.5, and 0.2 efficiency. Despite blocking similar antibodies, you would likely conclude that they behave differently due to the difference in x and z blocking. If this sounds like what you are looking for, I would start with help(hclust) in R and scroll down to the examples. Message: 23> Date: Sun, 15 Sep 2013 15:33:28 -0600 > From: Joshua Eckman <josheckman@hotmail.com> > To: "r-help@r-project.org" <r-help@r-project.org> > Subject: [R] hclust/dendrogram merging > Message-ID: <BAY167-W1059032665F8387DD05CB26DE250@phx.gbl> > Content-Type: text/plain > > I am working with protein blocking assays and the end result is a 2D > matrix describing which antibodies block the binding of other antibodies to > the target antigen.I need to group the antibodies together into "bins" > based on their combined profiles in both the row and column direction.I am > able to group the blocking profiles of rows vs rows, or columns vs columns, > using clustering. The end results could look something like this: > >col_bins binAb1 1Ab2 2Ab3 2Ab4 2Ab5 3Ab6 4Ab7 5Ab8 > 5Ab9 6 > In this case the "bin" values are just to describe they have similar > blocking profiles - so Ab2, Ab3, Ab4 have the same blocking profile, as do > Ab7 and Ab8. > Looking at the row profiles > >row_bins binAb1 1Ab2 2Ab3 3Ab4 3Ab5 4Ab6 5Ab7 5 Ab8 > 6Ab10 7 > The important end result, where I am stuck, is how to combine this with > the row direction and only report those that are represented in both > directions AND group together in both directions. It is possible that some > Abs will not be represented in both directions. The "bin" values of > row_bins and col_bins are also not important, just the relationship between > Abs by name that belong in the same bin, in both directions. > In other words, a combined bins report would look something like this: > binAb1 A Ab3 BAb4 BAb5 C > I made this visually because it is clear that these are the only groupings > that are maintained in both directions. But real data sets are much > bigger, so I need some form of automation. > Any ideas on how do this with matrix, dendograms or clustering functions? > Thank you, > josh >[[alternative HTML version deleted]]
Joshua, I'm not sure I understand your aim correctly, but if I do, here's my advice: If you are able to find the clusters according to rows or columns using clustering, you must be using some kind of a distance matrix that encodes whether two antibodies should be in one bin for rows, and a similar matrix for the columns. To get a clustering that represents only bins that occur in both directions, you can appropriately combine the two matrices into a single matrix. For example, if the distance matrix is zero if the antibodies go together and 1 otherwise, you can add the two matrices into a single matrix, then cluster the antibodies using the combined matrix using hclust (with complete linkage, if I understand it correctly), then use cutree() with cut height equal say 0.5. HTH, Peter On Sun, Sep 15, 2013 at 11:33 PM, Joshua Eckman <josheckman at hotmail.com> wrote:> I am working with protein blocking assays and the end result is a 2D matrix describing which antibodies block the binding of other antibodies to the target antigen.I need to group the antibodies together into "bins" based on their combined profiles in both the row and column direction.I am able to group the blocking profiles of rows vs rows, or columns vs columns, using clustering. The end results could look something like this: >>col_bins binAb1 1Ab2 2Ab3 2Ab4 2Ab5 3Ab6 4Ab7 5Ab8 5Ab9 6 > In this case the "bin" values are just to describe they have similar blocking profiles - so Ab2, Ab3, Ab4 have the same blocking profile, as do Ab7 and Ab8. > Looking at the row profiles >>row_bins binAb1 1Ab2 2Ab3 3Ab4 3Ab5 4Ab6 5Ab7 5 Ab8 6Ab10 7 > The important end result, where I am stuck, is how to combine this with the row direction and only report those that are represented in both directions AND group together in both directions. It is possible that some Abs will not be represented in both directions. The "bin" values of row_bins and col_bins are also not important, just the relationship between Abs by name that belong in the same bin, in both directions. > In other words, a combined bins report would look something like this: > binAb1 A Ab3 BAb4 BAb5 C > I made this visually because it is clear that these are the only groupings that are maintained in both directions. But real data sets are much bigger, so I need some form of automation. > Any ideas on how do this with matrix, dendograms or clustering functions? > Thank you, > josh > > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.