Ion-Pot and Ion-Curr: Ion-Aware Benchmarks for Electrostatic Field and Observable Prediction in Complex Geometries

Electrostatic interactions and ionic transport are central to biomolecular function, but continuum solvers for these problems remain costly at scale. Recent learning-based surrogates have shown promise, yet most existing settings focus on fixed ionic conditions and pay limited attention to the role of mobile ions. We introduce two complementary benchmarks for ion-aware biomolecular surrogate modeling: Ion-Pot, for electrostatic potential prediction around biomolecules under varying ionic conditions, and Ion-Curr, for ionic current prediction in complex nanopore-based biomolecular systems. Together, these tasks span two related prediction regimes: high-dimensional field reconstruction and observable-level transport prediction. We further develop voxel-based convolutional models tailored to each setting and show that they provide accurate and efficient surrogates for both tasks. Across the benchmarks, explicit conditioning on ion concentration improves predictive performance over no-ion-input baselines, highlighting the importance of ion-aware learning. Notably, the ionic-current task can be modeled accurately with a lightweight (39K) architecture, suggesting that surrogate complexity may depend strongly on the dimensionality and structure of the prediction target. These results establish Ion-Pot and Ion-Curr as a paired testbed for studying ion-dependent surrogate modeling in biomolecular systems.