If the results are often there in the simulation of interactions between proteins, AI shows its limits when dealing with small drug molecules. David Baker, 2024 Nobel laureate in chemistry, highlights these weaknesses and warns that not all of these systems are yet fully developed. To understand his warning, let’s imagine a lock and its key. In the world of proteins, it’s not enough for the key to fit the lock – what scientists call bonding. It still has to trigger the right mechanism.
A protein can perfectly attach to its target (like a drug attaching to a cancerous protein) without producing the expected therapeutic effect. It can even have the opposite effect: instead of blocking harmful activity, it could activate it. Artificial intelligence, despite its prowess in predicting structures and connections, does not yet know how to anticipate these functional subtleties.
It’s like having an excellent locksmith who can make a key that fits perfectly into the lock, but can’t tell you if that key will open the door, lock it permanently, or set off an alarm. In the medical context, this uncertainty is problematic for the design of new drugs. Current algorithms therefore excel in the art of “physical contact” between molecules, but remain blind to the complex choreography of biological interactions that ensues.
Health
