In Curious Wavefunction's field, chemistry (which is something I am, uh, somewhat familiar with), graph theory could probably be useful, but useful graph theoretic construction for retrosynthetic analysis seems to be persistently elusive since at least E. J. Corey's era.
But sometimes just being able to do math is important. Today I got into a minor dispute with someone who is running a sequencing/diagnostics startup. Their sequencing technique is, reported as of late 2015, getting an 8% error rate for an 8x coverage sequence. Asked him what the error would have been for a 1x coverage, and he couldn't answer. I would think that if you're going to be using sequencing to do diagnostics, being able to rapidly figure out things like this is important.