I've worked in bio and drug discovery for some 25 years. That includes building classifiers using gradient descent in the 90s (when algorithms, computers and data were all much worse). I ported DOCK to Linux in ~96 or 97. Since then I built an academic and then industrial career with some emphasis on using computing to solve problems in drug discovery, but I don't play that role any more.
It doesn't look like the models produced by this would immediately turn the challenging problem of finding, approving, and marketing successful pharmaceuticals (IE, it doesn't eliminate any real bottleneck).
There was a long-term dream of structure-based drug discovery based on docking, but IMO, it has never really proved itself (most of the examples of success are cherry picked from a much larger pile of massive failures).
It doesn't look like the models produced by this would immediately turn the challenging problem of finding, approving, and marketing successful pharmaceuticals (IE, it doesn't eliminate any real bottleneck).
There was a long-term dream of structure-based drug discovery based on docking, but IMO, it has never really proved itself (most of the examples of success are cherry picked from a much larger pile of massive failures).