Thread by @RyanLCollins13

Careful benchmarking of ~70 different methods for SV detection from human genomes:
genomebiology.biomedcentral.com/articles/10.11…

Looks to be an impressively comprehensive and valuable reference to guide future SV discovery efforts.

Thank you to the authors for compiling this study!!

FWIW, our own conclusions from other data broadly agree with this analysis:

1. No single algorithm is sufficient
2. Integrating multiple algs is critical
3. Some combos have better synergy than others
4. Mixing evidence types (e.g., one PE/SR alg, one depth alg, etc) is best

That said, if you were restricted to one or a handful of callers, we again agree w/authors:

- Manta is very good (perhaps best) all-purpose alg

- MELT (from @EugenomeUK) is currently the best-in-class MEI caller

One last point to emphasize: neglect depth-based algs at your own peril!

Although contribution to total # of variants seems small, RD algs disproportionately capture large & repeat-mediated CNVs

Absolutely critical to use read depth-based discovery for comprehensive SV calling

For more reading that is complementary to (and mostly in agreement w/) this benchmarking study, check out:

nature.com/articles/s4158… (tons of data in supplement)
nature.com/articles/s4146… (especially Fig 4)
biorxiv.org/content/10.110…

E.g., 👇 is Fig 4 from the HGSV paper from @mjpchaisson / Ashley Sanders / @Xuefang6 + large team. Makes the same points as above: