Symbolic Neural Generation with Applications to Lead Discovery in Drug Design
Researchers have developed a novel neurosymbolic model called Symbolic Neural Generators (SNGs) that combines Inductive Logic Programming with large language models. These SNGs learn from a small set of data instances to generate new, valid data that adheres to formal correctness criteria. The system produces both a symbolic description of feasible instances and a set of generated new instances. Initial evaluations in early-stage drug design show SNGs performing comparably to state-of-the-art methods on benchmark problems and generating molecules with binding affinities on par with leading clinical candidates for exploratory problems. AI
IMPACT This neurosymbolic approach could accelerate drug discovery by generating novel molecular candidates with high binding affinities.