Constructor University and Constructor Labs introduced BiteNetI, a deep-learning model designed to map ion binding sites directly onto three-dimensional protein structures. The tool, described as open-access, identifies binding sites for 14 ion types in seconds per structure and reportedly improves prediction accuracy by two- to threefold versus many existing methods. The model’s performance and speed could support downstream formulation and drug discovery tasks where binding-site characterization impacts protein function, stability and interaction profiles. The approach was published in Communications Biology. For the industry, the near-term use case will likely be speeding target and mechanism interpretation, particularly when experimental binding characterization is slow or when proteins adopt multiple conformational states across environments.