ForceFlowAb: physics-aware mixture-of-experts flow matching model for antibody CDRs design
Abstract Motivation: Antibodies are a major class of therapeutic molecules, and their recognition of target antigens is largely mediated by complementarity-determining regions (CDRs), making antigen-conditioned CDR design a central problem in antibody engineering. Recent generative methods have enabled antigen-conditioned co-design of CDR sequences and structures, but their limited capacity to…
Antibodies are crucial therapeutic molecules, with their target antigen recognition primarily driven by complementarity-determining regions (CDRs). CDRs are the focus of attention in antibody engineering, as they determine the binding specificity and affinity. Recent generative methods have enabled the co-design of CDR sequences and structures, tailored to a specific antigen.
However, these methods struggle to capture local interface heterogeneity and do not have explicit energy-based guidance during sampling, leading to suboptimal antibody-antigen interaction energies.
To address these limitations, researchers have developed ForceFlowAb, a physics-aware mixture-of-experts flow-matching model for antigen-conditioned CDR sequence-structure co-design. ForceFlowAb incorporates an adaptive routing mechanism that models heterogeneous interface environments, enabling the system to handle diverse binding modes. The model also applies differentiable force-field guidance during sampling, steering the CDR generation towards energetically favorable conformations.
When compared to existing methods such as FlowDesign and Diffab, ForceFlowAb demonstrates superior performance. For CDR-H3 design, ForceFlowAb achieved more favorable antibody-antigen interaction energies compared to both FlowDesign and Diffab, with improvement rates of 46.5% and 35.0% and 35.5%, respectively. In simultaneous six-CDR design, ForceFlowAb outperformed Diffab, with IMP values of 16% against 9%.
These results indicate that ForceFlowAb combines the benefits of interface-adaptive modeling and energy-based guidance, capturing binding-mode diversity while ensuring biophysical feasibility through physical constraints.
The ForceFlowAb web server is freely available at http://zhanglab-bioinf.com/ForceFlowAb, and the source code and implementation are accessible at https://github.com/iobio-zjut/ForceFlowAb. For further inquiries, contact can be made at zgj@zjut.edu.cn. Supplementary data are also available at Bioinformatics online.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
