The problem. DNA aligners like BWA assume a read maps to one contiguous stretch of genome. RNA reads don’t: a read can straddle a splice junction, with its two halves landing on exons thousands of bases apart. Force a contiguous aligner onto that and you either miss junction-spanning reads or misplace them.

The idea. STAR is built for spliced alignment. It uses a suffix-array-based search to find maximal mappable seeds within a read, then stitches seeds across introns — discovering splice junctions in the process rather than requiring them up front. It’s aggressively engineered for speed, aligning reads far faster than earlier spliced aligners at the cost of a large in-memory index.

Why it matters. STAR (or a pseudo-aligner) is the mapping step that produces the counts my aml_rnaseq_nf pipeline feeds to DESeq2. Reading it clarifies that RNA alignment is a genuinely different problem from DNA alignment — the splice-awareness is the whole point — and why the choice of aligner shapes what junctions I can even see. It’s the RNA counterpart to the BWA paper from yesterday.

Verdict. Foundational and a default to this day; HISAT2 and pseudo-aligners (kallisto, Salmon — coming up) offer different speed/memory trade-offs. Read it as the splice-aware half of the alignment story.