AIMS-Fold: Integrating Structural Proteomics with Diffusion Models for Protein Complex Conformation
The new AIMS-Fold framework has been launched to bridge the gap between generative protein structure models and the data from structural proteomics. While many current models predict static structures based on sequences, they often overlook the accurate conformational states of protein complexes, which are crucial for designing proteins, including PROTACs and antibodies. Techniques such as Cross-Linking Mass Spectrometry (XL-MS) and Hydrogen-Deuterium Exchange (HDX-MS) provide important spatial and dynamic insights, but the challenge lies in integrating these diverse measurements into generative models. AIMS-Fold employs a guided-diffusion method at inference time, steering generative samplings with physical potentials derived from XL-MS and HDX-MS data. The results can be found in a preprint on arXiv (2605.26192).
Key facts
- AIMS-Fold is an inference-time guided-diffusion framework.
- It integrates XL-MS and HDX-MS data into protein structure generative models.
- Current models fail to capture correct conformational states of protein complexes.
- The method uses differentiable physical potentials from XL-MS and HDX-MS.
- It combines structural proteomics data with pretrained diffusion models.
- The work is relevant for protein design and induced proximity modalities.
- The preprint is available on arXiv with ID 2605.26192.
- The approach addresses an open challenge in integrating heterogeneous measurements.
Entities
Institutions
- arXiv