We offer simple functions, in Python and R, which provide programmatic access to remote computation and remote data. The microservices architecture is described by Martin Fowler. He argues that software systems are most flexibly, robustly and adaptively constructed via composition, in which each of the component parts performs a single, well-defined task.
Current services include:
-
getDNAService: currently a lightweight wrapper around an established, public UCSC DAS server. The reference sequence for the specified genome, and chromosomal location is returned, reverse-complemented if requested.
-
fimoService: Invoke the MEME Suite's FIMO application, which scores the match of transcription factor binding motifs to sequences you submit. The FIMO client runs on your computer, the FIMO service (application) runs on whovian, an ISB linux server visible only within the firewall. The target motifs, expressed as PWMs, is obtained from JASPAR with some additions. Client/server communication is accomplished with the ZMQ protocol, the same one used by Jupyter notebooks.
These first two services are complementary: the getDNAService provides the DNA sequence - typically short --
which the fimoService requires. Once the software is installed (see each respective package for language-specific instructions),
these operations become possible, as demonstrated here with the loss of binding motifs due to a
SNP rs146894928, chr1:172883225 C/T on forward strand.
These coordinates are hg38. (A programmatic "liftover" service would come in handy.)
library(getDNAClient)
dnaClient<- getDNAClient("hg38")
start <- 172883225 - 5
end <- 172883225 + 5
seq.wt <- getSequenceByLoc(dnaClient, "chr1", start, end)
seq.snp <- sprintf("%s%s%s", substr(seq.wt, 1, 5), "T", substr(seq.wt, 7, 11))
We now have the wild-type and variant sequence. Submit each to fimo. Does this SNP affect predicted TF binding? Note that requestMatch requires that the sequences you give it be contained in an R list. In the two calls shown below, the lists are named. If you pass in an unnamed list, names are automatically generated for you.
fimo <- FimoClient("whovian", 5558)
print(requestMatch(fimo, list(wt=seq.wt))) # an empty table
print(requestMatch(fimo, list(mut=seq.snp))) # four motifs reported
Note, also, that you can combine sequences into a single named list, and make just one call:
print(requestMatch(fimo, list(wt=seq.wt, mut=seq.snp))) # still just four motifs reported