sex-label-check

Check for sex mislabels

Sex mislabeling implies sample mislabels, so here we’re looking for samples which group with the opposite of their sex label. Sorry libs, this is quite binary.

It happens to be the case that using the top two PCs calculated on the top 500 most variable genes very cleanly separates the samples by sex. We use this to do this analysis below:

pca_sex_full = suspicious_samples_data %>%
  filter(purpose == 'experiment') %>%
    ggplot(aes(PC1, PC2, 
               color = sex,
               key = subject,
               size = inferred_sex_mislabel,
               alpha = inferred_sex_mislabel)) +
      geom_point() +
      geom_abline(slope=slope,
                  intercept = intercept) +
      scale_alpha_manual(values = c('TRUE' = 1, 'FALSE' = 0.1)) +
      scale_size_manual(values = c('TRUE' = 5, 'FALSE' = 1)) +
      scale_color_manual(values = c("1" = "#B3589A",
                                    "2"="#9BBF85",
                                    "0" = "#F4A000"))

pca_sex_full

pca_sex_suspicious_labelled = suspicious_samples_data %>%
  filter(inferred_sex_mislabel, sex != 0) %>%
  ggplot(aes(PC1, PC2,
             color = sex,
             label = ifelse(inferred_sex_mislabel & sex != 0, 
                            paste(as.character(subject), 
                                  as.character(visit), 
                                  sep="_"), NA))) +
  geom_point(aes(size = inferred_sex_mislabel, alpha = inferred_sex_mislabel)) +
  geom_text_repel(color='black') +
  geom_abline(slope=slope,
              intercept = intercept) +
  scale_alpha_manual(values = c('TRUE' = 1, 'FALSE' = 0.1)) +
  scale_size_manual(values = c('TRUE' = 5, 'FALSE' = 1)) +
  scale_color_manual(values = c("1" = "#B3589A",
                                "2"="#9BBF85",
                                "control" = "#F4A000"))

pca_sex_suspicious_labelled

Update the Database

tl;dr: review the results, update the database, and update the dds object QC as appropriate.

Ideally, these sex mislabels are already captured in the database. Here, we check that these samples are all labelled “suspicious_sex”, that there are no “suspicious_sex” samples that aren’t here (eg, maybe we were able to positively identify a sample and forgot to remove the suspicious_sex label), and update the database if necessary.

Editing the database with this information can be a challenge, and requires that you understand the table structure. Please read the documentation on the database, and get comfortable with the key relationships between tables, and the llfs_rnaseq_metadata view.

Except for new samples which have not been reviewed, all old samples which display unusual grouping should be accounted for in the suspicious_sex column of sample (this is TRUE when WGS data is unavailable), in mislabel (this is TRUE when the wgs data says the sample isn’t what it says it is, but there is no positive match), or in a sample table notes, which might state that the sample was deemed to likely be correctly labelled for some other reason.

If you notice something wonky here, it takes a good deal of investigation – possibly you just need to update new sample info, most likely the suspicious_sex boolean value – but you might also need to do some investigating to make sure you agree with how samples are being labelled and re-labelled.

sex 1 suspicious samples

sex_1_suspicious_samples = suspicious_samples_data %>%
  filter(inferred_sex_mislabel, inferred_sex == 1, sex !=0) %>%
  select(library_id, subject,visit,batch,sex,
         inferred_sex,suspicious_sex, mislabel,relabel,notes)

sex 2 suspicious samples

sex_2_suspicious_samples = suspicious_samples_data %>%
  filter(inferred_sex_mislabel, inferred_sex == 2, sex !=0) %>%
  select(library_id, subject,visit,batch,
         sex,inferred_sex,suspicious_sex,mislabel,
         relabel, notes)