Skip to content
No description, website, or topics provided.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data/series_lists
plots
results
.gitignore
01-sample_series.Rmd
01-sample_series.nb.html
02-get_probes_from_bioconductor_pkgs.Rmd
02-get_probes_from_bioconductor_pkgs.nb.html
03-calculate_identifiers_overlap.Rmd
03-calculate_identifiers_overlap.nb.html
04-plot_highlighting_individual_series.Rmd
04-plot_highlighting_individual_series.nb.html
LICENSE
README.md
beadarray-platform-detection.Rproj

README.md

Illumina BeadArray Platform Detection

We're interested in automatically detecting platform/annotation based on the Illumina expression beadchip data that is publicly available. Some background discussion here. This repository contains some preliminary analyses of human platforms.

Data and methodology overview

Series Lists

We're looking at Illumina Human "whole genome" chips. (These chips have additional transcripts beyond what are on the "Ref" platforms.)

These are the GEO accessions we're using (from the linked issue above):

Platform Name GEO Accession
Human-6 v1.0 GPL2507
Human-6 v2.0 GPL6102
HumanHT-12 v3.0 GPL6947
HumanHT-12 v4.0 GPL10558

data/series_lists from each of these platforms were obtained through the GEO Browser on May 17, 2018. The platform accession was used as a search term -> click on series -> Export -> All Search results & Tab.

Non-normalized data

For each platform, we randomly selected 30 series that had supplementary txt files (e.g., there was a chance that there was non-normalized.txt files were available). We downloaded supplementary files that matched this pattern: *non*. In this sample, newer platforms had more accessions with raw data (as defined by this pattern; see table below).

platform no. accessions
GPL2507 10
GPL6102 18
GPL6947 27
GPL10558 27

List of probes and calculating overlap

Lists of probes were obtained from the following bioconductor packages (v1.26.0): illuminaHumanv1.db, illuminaHumanv2.db, illuminaHumanv3.db, illuminaHumanv4.db

For each series, we calculated what proportion of identifiers were in each of the lists of probes and the proportion of probes that were in the identifiers. Identifiers were (naively) assumed to be in the first column, as this is generally consistent with GEO instructions.

Results

The newer platforms (v2 and beyond) have some overlap in their identifiers. However, for most series, the identifiers from the data had the highest overlap with the platform for which it was labeled. There were some exceptions. We highlight the findings from the series labeled GPL6947 (HumanHT-12 v3.0 ) below.


Fig 1. Overlap between identifiers from series labeled GPL6947 and the four platform Bioconductor packages.

For most series (shown in black), we see the highest overlap with Humanv3 and some amount of overlap with v2 and v4. This is consistent with what we would expect based on the relationships between the platforms themselves. However, some series contained multiple Illumina platforms (and were not SuperSeries), which we would not have detected given our methodology. Once we color points from different platforms, we see that they behave as expected. GSE34074 GPL10558 is consistent with the pattern of most GPL10558 (HumanHT-12 v4) series. The identifiers from GSE25580 GPL6104 (HumanRef-8 v2.0) have the highest overlap with Humanv2 and less than 50% of the probes from v2, v3, and v4 are in the series identifiers, consistent with the smaller subset of transcripts present on Ref chips.

Below, we describe series with overlaps that deviated from this pattern for other platforms.

GPL2507 (Human-6 v1.0)
  • GSE17241 contains multiple platforms. The GPL6106 raw data appears to use bead IDs rather than probe IDs.
GPL6102 (Human-6 v2.0)
  • GSE14295 uses gene symbols rather than probe IDs. We'd be unable to get the probe sequences, which are required for processing, for this experiment.
  • GSE35102 looks like it might be could be WG-6 v2 filtered to only probes that are also present on the v3 chip. This would not matter for processing, as we'd be able to obtain gene identifiers and probe sequences for all the probes.
  • GSE45331 contains a PROBE_ID column that is not the first column, so we missed it using this method.
GPL10558 (HumanHT-12 v4)
  • The non-normalized file for GSE62374 contains lines of trailing whitespace (e.g., NA that were not removed).
  • The first column in GSE39417 contains gene symbols.
  • GSE54661 may be a mislabeled and in fact be from a whole genome v3 platform.
You can’t perform that action at this time.