The problem. Each clue to a structural variant is partial on its own. Read pairs that map at the wrong distance tell you a break is nearby but not exactly where. Split reads pin the break to the base but are sparser and easier to get wrong. Use one signal alone and you either lose resolution or lose sensitivity.

The idea. DELLY uses both. It first groups abnormal read pairs to find the rough location and type of a rearrangement, then looks for split reads across that region to place the breakpoint at single-base resolution. Merging the two kinds of evidence gives calls that are both sensitive and precise, across deletions, duplications, inversions, and translocations.

Why it matters. This idea of combining two weak signals into one strong call is the core of short-read SV detection, and Manta uses the same pairing. Reading DELLY next to Manta shows the shared logic behind the tools, so I can pick between them for variant_calling_nf on reasons rather than habit. For AML, where rearrangements matter, a precise breakpoint is what makes a call clinically meaningful.

Verdict. A foundational SV caller whose integrated approach shaped the field. It is slower than newer tools and, like all short-read methods, struggles in repeats. Read it for the clear statement of why two signals beat one.