Quality Control Fundamentals

1. Verify Download Integrity, Then Run Read-level QC

Before interpreting sequence quality, verify that each downloaded FASTQ file matches the checksum manifest provided by the archive or sequencing facility. A provider manifest contains the expected checksum for each exact filename. Do not use a checksum file you created only after an uncertain download as evidence that the original transfer was complete.

# Use the algorithm supplied by the data provider.
# SHA-256 manifest example:
sha256sum -c SHA256SUMS | tee checksum_validation.log

# Common sequencing-archive MD5 manifest example:
# md5sum -c md5checksums.txt | tee checksum_validation.log

# Confirm each compressed FASTQ stream can be read.
gzip -t sample_R1.fastq.gz sample_R2.fastq.gz

Proceed only when every required file reports OK and gzip -t returns no error. Save the provider manifest and checksum_validation.log under logs/; re-download any file that fails rather than editing the manifest.

For short reads, then inspect per-base quality, adapter content, sequence length, GC distribution, overrepresented sequences, and duplication. A low-quality tail may be trimmed, but trimming should be justified and recorded.

fastqc sample_R1.fastq.gz sample_R2.fastq.gz
multiqc .

2. Alignment and coverage QC

Mapping rate, properly paired reads, insert size, coverage depth, duplicate fraction, and target enrichment describe different failure modes. A high mapping rate does not guarantee correct alignment if the reference is wrong or contamination is present.

samtools flagstat aligned.bam
samtools idxstats aligned.bam | head
samtools depth -a aligned.bam | awk "{sum+=\$3} END {print sum/NR}"

3. Single-cell QC

For scRNA-seq, inspect genes per cell, counts per cell, mitochondrial proportion, ribosomal content, doublets, and cell-cycle or stress signals. Thresholds should be explored by tissue and protocol, not copied blindly from another dataset.

# Inspect the distribution of core QC metrics
qc_columns = ["n_genes_by_counts", "total_counts", "pct_counts_mt"]
print(adata.obs[qc_columns].describe())

# Visualize distributions and metric relationships before choosing thresholds
sc.pl.violin(adata, ["n_genes_by_counts", "total_counts", "pct_counts_mt"], jitter=0.25, multi_panel=True)
sc.pl.scatter(adata, x="total_counts", y="pct_counts_mt")

library(Seurat)

# Add QC percentages; use ^mt- for many mouse annotations
seurat_obj[["percent.mt"]] <- PercentageFeatureSet(seurat_obj, pattern = "^MT-")
seurat_obj[["percent.rb"]] <- PercentageFeatureSet(seurat_obj, pattern = "^RP[SL]")

# Inspect distributions before defining any filtering rule
VlnPlot(seurat_obj,
        features = c("nFeature_RNA", "nCount_RNA", "percent.mt", "percent.rb"),
        ncol = 4, pt.size = 0.1)
FeatureScatter(seurat_obj, feature1 = "nCount_RNA", feature2 = "percent.mt")

4. Record decisions

Create a QC report that states the metric, threshold, number removed, reason, and whether the decision was made before examining the biological outcome.

metric,threshold,removed,reason
pct_counts_mt,<20,143,high mitochondrial content

Practical Exercise

Choose one assay and create a one-page QC decision table with metric, plot, threshold rationale, records removed, and possible biological bias.

Pass criteria: You can explain at least four QC metrics, state why each matters, and document an exclusion without claiming that one universal threshold is correct.

Troubleshooting

If a sample fails every metric, do not rescue it by repeatedly changing cutoffs. Check sample identity, library preparation, contamination, sequencing depth, and batch before deciding.

Next Steps

Continue with Statistics for Bioinformatics and scRNA-seq Basics. Record the software versions, dataset or example inputs, and any decisions you made.