AbFlowNet: Optimizing Antibody-Antigen Binding Energy via Diffusion-GFlowNet Fusion

TL;DR

AbFlowNet presents a novel fusion of diffusion models and GFlowNets to directly optimize antibody–antigen binding energy during de novo CDR design. By reframing the diffusion process as a GFlowNet and applying a Trajectory Balance objective, it jointly optimizes reconstruction quality and energy rewards, avoiding the costly online RL loop and unreliable energy estimators. Empirical results show improvements in amino-acid recovery and structural fidelity, plus meaningful gains in binding-energy metrics compared to a diffusion baseline and competitive performance against RL-based methods, without using test-set complexes for training. The approach offers a scalable, energy-aware design paradigm that can leverage precomputed energy signals and reduces data leakage concerns while maintaining diffusion-model flexibility for diverse CDR generation.

Abstract

Complementarity Determining Regions (CDRs) are critical segments of an antibody that facilitate binding to specific antigens. Current computational methods for CDR design utilize reconstruction losses and do not jointly optimize binding energy, a crucial metric for antibody efficacy. Rather, binding energy optimization is done through computationally expensive Online Reinforcement Learning (RL) pipelines rely heavily on unreliable binding energy estimators. In this paper, we propose AbFlowNet, a novel generative framework that integrates GFlowNet with Diffusion models. By framing each diffusion step as a state in the GFlowNet framework, AbFlowNet jointly optimizes standard diffusion losses and binding energy by directly incorporating energy signals into the training process, thereby unifying diffusion and reward optimization in a single procedure. Experimental results show that AbFlowNet outperforms the base diffusion model by 3.06% in amino acid recovery, 20.40% in geometric reconstruction (RMSD), and 3.60% in binding energy improvement ratio. ABFlowNet also decreases Top-1 total energy and binding energy errors by 24.8% and 38.1% without pseudo-labeling the test dataset or using computationally expensive online RL regimes.

Figures (5)

• Figure 1: AbFlowNet reframes the diffusion process as a GFlowNet where each partially denoised CDR is a state and the transition probabilities are flows through edges. The initial state's flow is learned and the final state's flow is the binding energy of the reference CDR. To train, we simply enforce forward and backward flow parity, in addition to the diffusion losses.
• Figure 2: Antibody-antigen complex.
• Figure 3: De novo Generated and Reference CDR-H3s for 5MES complex. For DiffAb and AbFlowNet, we generated 100 CDRs and selected the one with the highest $\Delta \text{G}$.
• Figure 4: Hyperparameter search for TB loss weight $w$ in Eqn. \ref{['eqn:loss_combo']} on the RAbD rabd dataset. The RMSD of DiffAb on L3 CDR region is significantly worse than AbFlowNet. We repeated the retrained DiffAb using a different seed to confirm this discrepancy (RMSD $4.06$ Å).
• Figure 5: Training Diffusion+GFlowNet models with different training steps on the RAbD dataset rabd. Separating the reconstruction and flow matching steps do not meaningfully improve performance over AbFlowNet.