A Simple yet Effective DDG Predictor is An Unsupervised Antibody Optimizer and Explainer

TL;DR

This work tackles the enormous combinatorial space of antibody mutations by proposing Light-DDG, a lightweight, structure-aware DDG predictor trained through data augmentation and knowledge distillation from a strong teacher. It then elevates the utility of this predictor with Mutation Explainer, which uses iterative Shapley-value estimation to learn mutation sites and per-site preferences, enabling efficient, preference-guided antibody optimization under the Uni-Anti framework. The approach achieves state-of-the-art DDG prediction performance, enables fast, high-quality antibody screening against SARS-CoV-2, and provides interpretable mutation trajectories. The combination of an efficient predictor and explainable optimization offers a practical route toward high-throughput, unsupervised protein evolution with explainable guidance."

Abstract

The proteins that exist today have been optimized over billions of years of natural evolution, during which nature creates random mutations and selects them. The discovery of functionally promising mutations is challenged by the limited evolutionary accessible regions, i.e., only a small region on the fitness landscape is beneficial. There have been numerous priors used to constrain protein evolution to regions of landscapes with high-fitness variants, among which the change in binding free energy (DDG) of protein complexes upon mutations is one of the most commonly used priors. However, the huge mutation space poses two challenges: (1) how to improve the efficiency of DDG prediction for fast mutation screening; and (2) how to explain mutation preferences and efficiently explore accessible evolutionary regions. To address these challenges, we propose a lightweight DDG predictor (Light-DDG), which adopts a structure-aware Transformer as the backbone and enhances it by knowledge distilled from existing powerful but computationally heavy DDG predictors. Additionally, we augmented, annotated, and released a large-scale dataset containing millions of mutation data for pre-training Light-DDG. We find that such a simple yet effective Light-DDG can serve as a good unsupervised antibody optimizer and explainer. For the target antibody, we propose a novel Mutation Explainer to learn mutation preferences, which accounts for the marginal benefit of each mutation per residue. To further explore accessible evolutionary regions, we conduct preference-guided antibody optimization and evaluate antibody candidates quickly using Light-DDG to identify desirable mutations.

Figures (7)

• Figure 1: Efficiency vs Effectiveness. There are three variants of Light-DDG with differing numbers of attention heads (default to 4 in this paper).
• Figure 2: (a) A binding energy landscape reflecting the mapping from mutations to $\Delta\Delta G$ scores. (b) Pre-training a Light-DDG with augmentation and distillation, and then using it as the core, together with mutation explainer and search, to construct a unified framework for antibody optimization.
• Figure 3: (a) Correlations between experimental and predicted $\Delta\Delta G$s. (b) Distributions of per-structure Pearson scores. (c) Per-structure Pearson correlation scores for eight complex examples.
• Figure 4: (a) Applicability of using different $\Delta\Delta G$ predictors as teachers. (b) Sensitivity to the sizes of augmentation data. (c) Robustness of different $\Delta\Delta G$ predictors to 3D structure Gaussian noise.
• Figure 5: Rankings of the five favorable mutations on the antibody screening against SARS-CoV-2.