TerraBind: Fast and Accurate Binding Affinity Prediction through Coarse Structural Representations

We present TerraBind, a foundation model for protein-ligand structure and binding affinity prediction that achieves 26$\times$ faster inference than state-of-the-art methods while improving affinity prediction accuracy by up to 20\%. Current deep learning approaches to structure-based drug design rely on expensive all-atom diffusion to generate 3D coordinates, creating inference bottlenecks that render large-scale compound screening computationally intractable. We challenge this paradigm with the hypothesis: full all-atom resolution is unnecessary for accurate small molecule pose and binding affinity prediction. TerraBind tests this hypothesis through a coarse pocket-level representation (protein C$_\beta$ atoms and ligand heavy atoms only) within a multimodal architecture combining pretrained molecular encoders and ESM-2 protein embeddings that learns rich structural representations, which are used in a diffusion-free optimization module for pose generation and a binding affinity likelihood prediction module. On structure prediction benchmarks, TerraBind matches diffusion-based baselines in ligand pose accuracy. For binding affinity, TerraBind outperforms Boltz-2 by 16-20\% in Pearson correlation on both a public benchmark (CASP16) and a diverse private dataset (18 assays). The affinity module also provides well-calibrated uncertainty estimates, addressing a critical gap in compound prioritization for drug discovery. Furthermore, this module enables a continual learning framework and a hedged batch selection strategy that, in simulated drug discovery cycles, achieves 6$\times$ greater affinity improvement over greedy approaches.

Developing new medicines requires finding chemical molecules that bind tightly to disease-causing proteins. Modern AI models can predict this binding with high accuracy, but they are painfully slow — they simulate the position of every atom step by step, taking about twenty seconds per molecule. This creates a major bottleneck when scientists need to screen the billions of compounds available in modern chemical libraries. We developed TerraBind, a streamlined AI model that eliminates this bottleneck. We found that simulating every atom in microscopic detail is unnecessary; instead, our model focuses on a simplified, high-level map of the crucial contact areas between the drug and the protein. By stripping away the expensive atom-by-atom step, TerraBind predicts binding 26 times faster than the leading method (Boltz-2) — and is also 16–20% more accurate on both a public benchmark and eighteen real drug discovery programs. TerraBind also provides a calibrated confidence score for each prediction, so scientists can prioritize the most reliable candidates and avoid wasting expensive lab experiments on uncertain guesses. By combining speed, accuracy, and trustworthy uncertainty estimates, TerraBind makes it practical to apply state-of-the-art AI to far larger chemical libraries than before — helping accelerate the search for treatments for diseases that currently lack good options.