counts <- readr::read_csv("data/demo-counts.csv", show_col_types = FALSE)
dim(counts)[1] 500 13Data Intake and Quality Control
Why QC at the count level matters
Before modeling, we must confirm that the count matrix is structurally sound.
Errors at this stage propagate silently into downstream analyses.
Quality control at this level focuses on:
- dimensional integrity
- sample-level sequencing depth
- count distribution structure
- low-information genes
Load the count matrix
The first column should be gene_id.
Remaining columns correspond to samples.
Confirm numeric counts
all(sapply(counts[-1], is.numeric))[1] TRUEAll sample columns must be numeric.
Library size assessment
Library size is the total number of reads per sample.
library_sizes <- colSums(counts[-1])
library_sizessample-01 sample-02 sample-03 sample-04 sample-05 sample-06 sample-07 sample-08
7684 8351 13817 9176 9442 14177 12318 7689
sample-09 sample-10 sample-11 sample-12
8618 9841 16535 12419 Visualize library sizes
library_df <- tibble::tibble(
sample = names(library_sizes),
library_size = library_sizes
)
ggplot2::ggplot(library_df, ggplot2::aes(x = sample, y = library_size)) +
ggplot2::geom_col(fill = "#036281") +
ggplot2::labs(
title = "Library Size per Sample",
x = "Sample",
y = "Total Counts"
) +
ggplot2::theme_light() +
ggplot2::theme(
plot.title = ggplot2::element_text(hjust = 0.5, face = "bold"),
plot.subtitle = ggplot2::element_text(hjust = 0.5),
panel.grid.minor = ggplot2::element_blank(),
legend.position = "right",
axis.text.x = ggplot2::element_text(angle = 45, hjust = 1)
)
Moderate variation is expected.
Extreme outliers require investigation.
Distribution of raw counts
RNA-seq counts are typically right-skewed.
raw_values <- as.vector(as.matrix(counts[-1]))
ggplot2::ggplot(
data.frame(count = raw_values),
ggplot2::aes(x = count)
) +
ggplot2::geom_histogram(bins = 50, fill = "#036281") +
ggplot2::scale_x_continuous(trans = "log10") +
ggplot2::labs(
title = "Distribution of Raw Counts (log scale)",
x = "Count (log10 scale)",
y = "Frequency"
) +
ggplot2::theme_light() +
ggplot2::theme(
plot.title = ggplot2::element_text(hjust = 0.5, face = "bold"),
plot.subtitle = ggplot2::element_text(hjust = 0.5),
panel.grid.minor = ggplot2::element_blank(),
legend.position = "right"
)Warning in ggplot2::scale_x_continuous(trans = "log10"): log-10 transformation
introduced infinite values.Warning: Removed 108 rows containing non-finite outside the scale range
(`stat_bin()`).
The heavy right tail reflects a small number of highly expressed genes.
Log transformation for visualization
log_counts <- log2(counts[-1] + 1)log_df <- tidyr::pivot_longer(
data.frame(sample = colnames(log_counts), t(log_counts)),
cols = -sample,
names_to = "gene",
values_to = "log_expression"
)
ggplot2::ggplot(log_df, ggplot2::aes(x = sample, y = log_expression)) +
ggplot2::geom_boxplot(fill = "#036281", outlier.size = 0.3) +
ggplot2::labs(
title = "Log2 Expression Distribution per Sample",
x = "Sample",
y = "Log2(count + 1)"
) +
ggplot2::theme_light() +
ggplot2::theme(
plot.title = ggplot2::element_text(hjust = 0.5, face = "bold"),
plot.subtitle = ggplot2::element_text(hjust = 0.5),
panel.grid.minor = ggplot2::element_blank(),
legend.position = "right",
axis.text.x = ggplot2::element_text(angle = 45, hjust = 1)
)
After log transformation, distributions should be more comparable.
Low-count filtering
Genes with extremely low counts across all samples contribute little information.
min_count_threshold <- 10
keep_genes <- rowSums(counts[-1] >= min_count_threshold) >= 3
sum(keep_genes)[1] 430This keeps genes expressed at or above the threshold in at least three samples.
filtered_counts <- counts[keep_genes, ]
dim(filtered_counts)[1] 430 13Filtering reduces noise and improves downstream stability.
Interpretation
Quality control at the count level ensures:
- sample-level integrity
- reasonable depth variation
- appropriate distributional structure
- removal of low-information features
No modeling is performed yet.
We are preparing the matrix for meaningful normalization and exploratory analysis.