“AI is good at interpolating features within the
bounds of a
well-explored pool of training data but unable to make useful
extrapolations far outside it."
“Because the availability of sufficient training
data is a serious
limitation, the … question then becomes how to get around this
experimental data bottleneck …. Properly integrating AI and molecular
modeling is likely to produce far more powerful breakthroughs in small
molecule drug discovery than either approach could individually.”