A Synthetic Benchmark for Topology-Aware Modeling of Multispecific Antibody Function

Multispecific antibodies (msAbs) offer transformative therapeutic potential by engaging multiple epitopes simultaneously, yet their efficacy is an emergent property governed by complex molecular architectures. Rational design is often bottlenecked by the inability to predict how subtle changes in domain topology influence functional outcomes, a challenge exacerbated by the scarcity of comprehensive experimental data. Here, we introduce Synapse, a biologically inspired synthetic benchmark for topology-aware modeling of msAbs. First, we present a generative method that combines nonlinear per-domain sequence-to-score landscapes with connectivity-dependent synthetic readouts. Second, we propose a graph neural network architecture that explicitly encodes these topological constraints, distinguishing between format configurations that appear identical to the evaluated connectivity-agnostic multilayer perceptron. We demonstrate that this model learns topology-dependent structure-function relationships on synthetic landscapes and evaluate a transfer learning workflow as infrastructure for adaptation to biological datasets. We showcase our framework with a trispecific T cell engager-inspired synthetic topology-dependent optimization task and a constrained ranking of common light chain candidates against a composite objective. This work provides a controlled benchmarking environment for developing and evaluating topology-aware computational tools for msAb modeling. All experiments use synthetic data; no biological validation is performed.