Out of some little experience doing some machine learning genetic research that's a lot similar to this study, I feel like the effect of such studies is merely directing us towards finding out the functionality of some parts of DNA that we don't know about.

But the study is having a low number of samples for a machine learning algorithm to yield perfectly accurate results, so I'd say any study with machine learning algorithms without a big enough sample, the results should be always taken with a bit of skepticism. There's no real substitute for good old biology research to know the exact functionality of genes, proteins ... etc