4.5 Infrastructure Discussion
We have developed the infrastructure to embed a return address in analysis workspaces that will enforce gate names and make it possible to collect the larger numbers of reference analyses needed to establish the variance within the gold standard from the collaborator's own technicians. This step forward will simplify the collection of the acceptable analysis for each use case based on collaborator's/contributor's own technician's analysis rather than Tree Star Application Scientists that do not have as much experience with each particular assay as the investigators that have been working intimately with the experiments. We will simply send a FlowJo workspace that references our database and FCS files and the technician will gate using specified gate names. The Save operation will send the analyzed workspace back to Tree Star and will load the database.
We are working within the flow cytometry community accepted standards to allow outside classification methods from R or others to plug into our database by using accepted standards such as GatingML. This will facilitate additional collaborations and support for projects such as FlowCAP.
During the past two years there has been increased academic interest, especially in the bioinformatics community. The Bioconductor project has brought skills and tools developed for microarray analysis into the world of Flow. Increased references appear on Purdue list, and the preprints and experiences we have of groups working in R is growing. To accommodate this trend we adapted our experimental and repository design to accept external classifiers. This has given us ways that we can bring a larger set of collaborators into the project moving forward, without making them follow some API to embed classification inside the FCS format. We have also developed the infrastructure to embed a return address in analysis workspaces that will enforce gate names and make it possible to collect the larger numbers of reference analyses needed to establish the variance within the gold standard.
We do have some limitations in our database design such that we can only compare populations where the name is strictly enforced and we only compared specified populations as defined by the assay set up. i.e. GvHD use case data, compare only CD4+, CD8b+ populations across algorithms. If we were to try to compare other clustering algorithms that search for any cluster, we would have to enforce the gate name to do the comparison to our data files. This is an acceptable limitation at this point because our focus is to have FlowDx as an automated solution for clearly defined clinical assays where the populations of interest and the logic to find these populations have been defined. We are not planning to have FlowDx solutions for exploratory flow cytometry research trying to find new populations.