This notebook explores the table of gene conversions a bit, to get a +bit of a sense of where different references might vary.
+The table includes the following reference data:
+We will focus on the ScPCA and 10x 2020 references for now, as few +data sets include the 2024 reference at this point. One thing to note is +that Ensembl 98 reference used BAC-based gene names for otherwise +unnamed genes, but as of Ensmbl 104 (the base for the ScPCA reference), +those were replaced with Ensembl ids.
+ + + +suppressPackageStartupMessages({
+ library(SingleCellExperiment)
+ library(dplyr)
+})Warning: replacing previous import ‘S4Arrays::makeNindexFromArrayViewport’ by ‘DelayedArray::makeNindexFromArrayViewport’ when loading ‘SummarizedExperiment’load(here::here("data", "scpca_gene_reference.rda"))We expect that many of the Ensembl IDs in the ScPCA data will not +have corresponding values in the 10x references, but some of the 10x +references have Ensembl IDs not present in the ScPCA data. Let’s look at +the number of missing gene symbols in the ScPCA data:
+ + + +scpca_missing <- scpca_gene_reference |>
+ filter(is.na(gene_symbol_scpca))How many of the gene symbols in the 10x 2020 reference are in the +ScPCA data, but perhaps with a different Ensembl ID?
+ + + +symbols_10x2020 <- scpca_missing$gene_symbol_10x2020[!is.na(scpca_missing$gene_symbol_10x2020)]
+sum(symbols_10x2020 %in% scpca_gene_reference$gene_symbol_scpca)[1] 10What are they?
+ + + +# Match the genes that are in the 10x 2020 reference to the ScPCA table on gene symbol
+# Only for those where the gene id is missing in the ScPCA table
+scpca_gene_reference |>
+ filter(
+ !is.na(gene_symbol_10x2020),
+ is.na(gene_symbol_scpca),
+ gene_symbol_10x2020 %in% scpca_gene_reference$gene_symbol_scpca
+ ) |>
+ select(gene_ids, gene_symbol_10x2020, gene_symbol_10x2020_unique) |>
+ left_join(
+ scpca_gene_reference |> select(gene_ids, gene_symbol_scpca, gene_symbol_scpca_unique),
+ by = c("gene_symbol_10x2020" = "gene_symbol_scpca"),
+ suffix = c("_10x2020", "_scpca")
+ )Looks like these are all cases where the gene id has been updated. In
+only one case is this a gene where something was not unique, which is
+LINC01505. This gene (ENSG00000234229) seems
+to have simply been removed in later revisions, so I think we can safely
+not worry about it.
scpca_gene_reference |>
+ filter(gene_symbol_10x2020 == "LINC01505")In general, simply translating to the list gene symbols should work +as expected.
+In Ensembl version 104, the version we are using for ScPCA, BAC-based +gene IDs were removed and replaced with simply the Ensembl gene ID. Many +of the gene symbols that are present in the 10x references but missing +in the ScPCA data are these BAC-based gene IDs. We will probably want to +translate these to the symbol used in the 10x reference when +requested.
+ + + +scpca_gene_reference |>
+ filter(
+ !is.na(gene_symbol_10x2020),
+ !gene_symbol_10x2020 %in% scpca_gene_reference$gene_symbol_scpca
+ )One other question is how often the unique gene symbols
+disagree for the same Ensembl ID. Here we will exclude the cases where
+the gene symbol is an Ensembl ID in the ScPCA data, as these are
+expected to be different.
scpca_gene_reference |>
+ filter(
+ gene_symbol_scpca != gene_ids,
+ gene_symbol_scpca_unique != gene_symbol_10x2020_unique
+ )More than I expected, but it looks like most of these are cases where +the gene symbol has been updated. We can verify this by looking at how +often the base symbol is the same.
+ + + +scpca_gene_reference |>
+ filter(
+ gene_symbol_scpca != gene_ids,
+ gene_symbol_scpca_unique != gene_symbol_10x2020_unique,
+ ) |>
+ mutate(same_symbol = gene_symbol_scpca == gene_symbol_10x2020) |>
+ count(same_symbol)What we are really interested in are the cases where the process of +making the gene symbol unique had different effects:
+ + + +scpca_gene_reference |>
+ filter(
+ gene_symbol_scpca == gene_symbol_10x2020,
+ gene_symbol_scpca_unique != gene_symbol_10x2020_unique
+ )This definitely happens more than I would have hoped, though my hope +was probably unreasonable, as 14 cases is not bad. Which is to say that +the best way to handle this is probably simply to translate using the +the table directly when performing comparisons to existing data for the +most accurate results.
+We will need to have clear instructions about when to use which kind +of translation.
+ +