Cell-level QC

Will Macnair - Computational Sciences Center of Excellence, F Hoffmann-La Roche Ltd, Basel, Switzerland

March 23, 2026

Sample QC metrics

An overview of quality control (QC) metrics for all samples, both before and after filtering. The QC metrics include:

no. of cells : Total number of barcodes called as cells for each sample.
no. of UMIs : Sum of counts across all features for each cell with log-transformed values displayed.
no. of genes : Number of unique features with non-zero counts for each cell with log-transformed values displayed.
mito pct. : The proportion of reads that mapped to genes in the mitochondrial genome. High proportions are indicative of poor-quality cells (compromised membranes allow individual RNA molecules to escape while retaining mitochondria, leading to an increased relative abundance of mitochondrial transcript) or nuclei (incomplete removal of cytoplasm). Logit-transformed values are displayed as applying the logit (inverse logistic) transformation to mitochondrial proportions yields approximately Gaussian distributions.
spliced pct. : The proportion of spliced reads. In single-nuclei RNA sequencing, high spliced proportions may indicate inadequate removal of cytoplasmic material from the nuclei. Logit-transformed values are displayed as applying the logit (inverse logistic) transformation to spliced proportions results in approximately Gaussian distributions.

suppressWarnings({
  cat("### Pre-QC\n")
  print(plot_qc_ranges_marginals(qc_dt, b_lvls, qc_names, qc_lu, cuts_dt, batch_var))
  cat("\n\n")
  cat("### After light QC\n")
  print(plot_qc_ranges_marginals(kept_dt, b_lvls, qc_names, qc_lu, cuts_dt, batch_var))
  cat("\n\n")
})

Pre-QC

After light QC

QC metrics scatterplots

Pairwise relationships of QC metrics at individual cell level.

for (b in b_lvls) {
  qc_tmp      = qc_dt[ batch_var == b ]
  if (nrow(qc_tmp) == 1)
    next
  cat('### ', b, '\n')
  print(plot_qc_metric_scatter(qc_tmp, qc_names, qc_lu, cuts_dt[ batch_var == b ], b))
  cat('\n\n')
}

SAMN29101222

SAMN29101249

SAMN29101227

SAMN29101247

SAMN29101284

SAMN29101255

SAMN29101262

SAMN29101235

SAMN29101271

SAMN29101226

SAMN29101231

SAMN29101274

SAMN29101238

SAMN29101245

SAMN29101290

SAMN29101206

SAMN29101292

SAMN29101278

SAMN29101256

SAMN29101279

SAMN29101223

How many cells and samples retained?

Table summarizing the number of cells before and after qc filtering. Some samples may have been excluded from further analysis due to insufficient number of cells.

calc_qc_summary(qc_dt, kept_dt, cuts_dt, qc_lu, batch_var) %>% knitr::kable()

sample_id	excluded	N pre-QC	N post-QC	N excluded	pct. excluded	pct. excluded by no. of UMIs	pct. excluded by no. of genes	pct. excluded by mito. pct.	pct. excluded by spliced pct.
SAMN29101223	FALSE	1157	254	903	78.0	13.7	10.2	66.8	5.5
SAMN29101279	FALSE	3314	1223	2091	63.1	24.5	19.3	44.3	7.2
SAMN29101255	FALSE	6405	3359	3046	47.6	14.9	12.6	45.6	3.5
SAMN29101227	FALSE	10497	5613	4884	46.5	26.7	19.3	39.7	5.4
SAMN29101292	FALSE	5710	3756	1954	34.2	5.9	6.4	27.9	10.2
SAMN29101222	FALSE	6889	4885	2004	29.1	7.4	7.7	26.2	4.0
SAMN29101235	FALSE	7075	5026	2049	29.0	5.7	6.7	25.4	4.6
SAMN29101271	FALSE	6899	4953	1946	28.2	7.3	6.1	25.1	1.7
SAMN29101247	FALSE	3337	2445	892	26.7	7.0	7.3	24.4	3.4
SAMN29101256	FALSE	1959	1473	486	24.8	3.1	4.0	3.9	22.7
SAMN29101262	FALSE	7431	5692	1739	23.4	7.1	8.0	20.5	3.6
SAMN29101278	FALSE	2834	2187	647	22.8	8.3	6.2	1.5	16.4
SAMN29101284	FALSE	8422	6590	1832	21.8	6.9	6.9	17.0	4.8
SAMN29101249	FALSE	8779	7135	1644	18.7	5.8	6.2	15.0	4.4
SAMN29101206	FALSE	8748	7208	1540	17.6	7.1	8.1	13.6	3.1
SAMN29101290	FALSE	3842	3172	670	17.4	7.2	7.4	13.7	3.3
SAMN29101226	FALSE	7781	6550	1231	15.8	4.1	5.2	12.3	2.6
SAMN29101274	FALSE	7510	6439	1071	14.3	4.0	4.6	10.9	2.6
SAMN29101231	FALSE	7562	6513	1049	13.9	3.2	4.5	10.9	2.5
SAMN29101238	FALSE	6588	5682	906	13.8	3.5	5.3	10.8	2.8
SAMN29101245	FALSE	5702	4956	746	13.1	3.2	4.2	9.0	3.8

What were the reasons for excluding barcodes?

An upset plot is generated for each sample to visualize the reasons for barcode exclusion. The plot shows the overlap between different exclusion criteria, with vertical bars representing the size of each intersection. Plots are omitted for samples where no barcodes were excluded.

for (bb in b_lvls) {
  qc_tmp      = qc_dt[ batch_var == bb ]
  if (nrow(qc_tmp) == 1 | sum(qc_tmp$keep) == nrow(qc_tmp))
    next
  cat('### ', bb, '\n')
  suppressMessages( print(plot_upset_of_exclusions(qc_tmp, qc_names, qc_lu, cuts_dt)) )
  cat('\n\n')
}