Related Experiment Video
Updated: Aug 5, 2026
10:31
Antigenic Liposomes for Generation of Disease-specific Antibodies
Published on: October 25, 2018
Mixture diffusion model for multimodal antibody design
Vasanth Durvasula1, Tiara Natasha Binte Sayuti1, Jagath C Rajapakse1
1College of Computing and Data Science, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798, Singapore.
Briefings in Bioinformatics
|August 3, 2026
Summary
This study introduces a mixture diffusion model for antibody design, enabling exploration of diverse antibody conformations for improved antigen binding. This multimodal approach overcomes limitations of single-mode models, facilitating discovery of high-affinity binders.
Area of Science:
- Biotechnology
- Computational Biology
- Structural Biology
Background:
- Antibody design requires modeling flexible complementarity-determining region (CDR) loops for high-affinity antigen binding.
- Current diffusion models use unimodal distributions, limiting exploration of diverse antibody conformations.
Purpose of the Study:
- To introduce a mixture diffusion model for multimodal antibody design.
- To overcome limitations of single-mode diffusion models in exploring antibody conformational space.
Main Methods:
- Developed a denoising diffusion probabilistic model using mixture density parameterizations for positional and rotational updates.
- Experimented on antibody-antigen complexes from the Structural Antibody Database (SAbDab).
Main Results:
- Increasing mixture components improved model performance by capturing distinct CDR backbone conformations.
- Achieved competitive amino acid recovery and binding-affinity metrics.
- Maintained physically consistent antibody backbones.
Conclusions:
- Mixture-based diffusion modeling offers a practical approach for discovering high-quality antibody conformations.
- This method enables access to conformations previously inaccessible with conventional single-mode diffusion frameworks.
