The problem. Aligning reads to a single linear reference bakes in reference bias: reads carrying an alternate allele map worse, or to the wrong place, precisely where the biology is interesting. RNA-seq adds spliced alignment across introns. You want a fast aligner that knows about population variation and handles splicing without ballooning memory.

The idea. HISAT2 builds a graph FM-index that folds millions of known variants and haplotypes into the reference structure, so reads align against a graph of alternatives rather than one sequence. A hierarchical index (global plus many local indexes) keeps it fast and memory-light while supporting spliced alignment. The same machinery drives HISAT-genotype for HLA typing and DNA fingerprinting — alignment and genotyping from one graph.

Why it matters. This sits right next to STAR in my RNA-seq stack as the spliced-aligner alternative, and it carries the FM-index/Burrows–Wheeler lineage — the same indexing idea from the foundations reading — into a graph reference. The reference-bias problem it addresses is the same one that motivated graph pangenomes and DeepVariant’s re-examination of alignments. For anyone building a variant or expression pipeline, it’s a core design choice.

Verdict. A widely-used, memory-efficient graph aligner; the graph helps most where good variant catalogues exist, and splicing accuracy still depends on annotation. Read it as alignment that treats the reference as a population, not a single string.