AI and machine learning can achieve a lot in drug discovery, but a fundamental limitation often shows up when you push beyond known chemistry: the generation of truly novel, viable scaffolds. Most workflows learn from existing data and can reliably suggest candidates that improve properties within familiar chemical space. But as soon as you extrapolate into new regions, uncertainty rises quickly as predictions become less trustworthy with increased error rates. If the goal is to reach genuinely new, patentable structures beyond known actives, these approaches usually need to be complemented with additional methods and concepts to reach the required level of confidence.
In this webinar, we explore approaches for unlocking new chemical modalities and take a look at the pitfalls and myths surrounding de novo compound ideation.
