Brief

Researchers have developed ShallowBench, a new benchmark designed to evaluate generative AI models used in drug design, specifically focusing on targets with shallow binding pockets. Existing models perform poorly on these challenging targets, which are common in areas like oncology. ShallowBench, comprising 5,780 targets, aims to drive innovation in AI architectures and loss functions to improve drug discovery for historically difficult-to-target proteins. AI

Researchers have developed a new framework called Neural Scalable Symbolic Search (NS3) to address the challenge of complex query answering over knowledge graphs. Existing methods struggle with queries involving multiple free variables, often relying on less accurate marginal rankings. NS3 approximates joint rankings by first answering marginalized sub-queries, then merging variables into pruned domains controlled by a dynamic budget, and progressively reducing the query complexity. AI