Short videos (2.5–4 minutes) that explain the ideas and the maths behind bioinformatics analyses, one small step at a time, with a hands-on workshop that reproduces every calculation in R. Everything uses one real data set: mouse hearts after myocardial infarction (MI) or sham surgery, at day 1 and day 3 (ENA project PRJNA412001).
Foundations
How sequencing works, from the bench to the files: no background needed.
From Sanger to next-generation sequencing
Sequencing means copying DNA one letter at a time and watching each letter. Chain termination, capillary electrophoresis, and what changed with massively parallel sequencing.
Library preparation
How an RNA sample becomes a library: poly-A capture, fragmentation, cDNA synthesis, A-tailing and adapters. The anatomy of one library molecule: P5, primer sites, insert, index, P7.
Cluster generation: bridge amplification
On the flow cell, each library molecule lands, is copied, bends over into a bridge, and doubles again and again into a cluster of about a thousand identical copies.
Sequencing by synthesis
Dyed, blocked nucleotides: add, photograph, cut, one letter per cycle. Why quality falls along a read, how the index read sorts samples, and how read 2 reads the other end.
FASTQ files and quality scores
Four lines per read. The Phred score, Q = −10 log₁₀ P, stored as one ASCII character per letter, and a FastQC-style check of real reads.
Reference files: FASTA and GTF
Genome and transcriptome FASTA files, reading a header, the nine columns of a GTF file, and why the genome build and annotation release must match.
Alignment, SAM and BAM
Splice-aware alignment of RNA reads (a real read jumping a 406-letter intron), and the SAM fields that matter: FLAG, MAPQ and CIGAR.
The pipeline: from FASTQ to a count table
Quality control, alignment and counting, or pseudo-alignment; the count matrix and its metadata; how 30 GB of reads become a 1.4 MB table.
Transcriptomics
From a count table to statistics and pathways: bulk RNA-seq analysis, step by step.
What does RNA-seq measure?
Counts of mRNA fragments as a proxy for gene activity, the experiment used throughout, and the road map of the transcriptomics part.
From reads to counts
k-mers and pseudo-alignment as the intersection of compatibility sets, and the EM algorithm that shares out ambiguous reads.
Counts, lengths and TPM
Sequencing depth and gene length biases; TPM versus RPKM, worked by hand; why TPM is a share of a fixed total.
PCA, step by step
The line of maximum variance, built from scratch with two genes; then why raw TPM misleads and a variance-stabilising transformation helps.
Size factors: the median of ratios
DESeq2's size factors, computed by hand on five real genes: a geometric-mean pseudo-reference and the median ratio.
Counts are noisy: the negative binomial
Why counts vary more than Poisson predicts, variance = μ + αμ², and how DESeq2 shrinks dispersions towards a trend.
Testing one gene: the GLM and the Wald test
The generalised linear model behind DESeq2, the log fold change as β₁, and the Wald test z = β₁ / SE, worked on real genes.
20,000 tests at once: the false discovery rate
p-value histograms, Bonferroni and Benjamini–Hochberg by hand, and why some adjusted p-values are NA.
From gene lists to pathways
Over-representation with the hypergeometric test, gene set enrichment's running sum, piano, and why results need a sanity check.
Designs with more than one factor
Design formulas and model matrices, additive versus interaction models, and correcting for batch.