New tool enhancing Nobel Prize-winning AlphaFold
In the CASP15 experiment, AlphaFold's ability to predict multi-protein structures improved significantly, though it still faces challenges with large protein complexes and integrating experimental data. As part of a cross-platform Technology Developemnt Project, NBIS has been driving to address these limitations in close collaboration with the Cellular and Molecular Imaging and the Integrated Structural Biology platforms and researcher Björn Wallner (LiU), now resulting in a new tool called AF_unmasked, enhancing the accuracy of complex protein structure predictions by merging experimental data with computational models.
Claudio Mirabello, a bioinformatician at SciLifeLab’s NBIS platform, explains, “AlphaFold's predictions are impressive but limited without incorporating experimental data. AF_unmasked bridges this gap, marking the first step toward combining computational models with real biological observations.” AF_unmasked can produce highly reliable structures, even in difficult scenarios like limited evolutionary data or incomplete experimental models, with a DockQ score above 0.8, indicating strong accuracy.
A key feature of AF_unmasked is its ability to fill in gaps in experimental models and suggest multiple conformations, offering insights into protein movement and function. Marta Carroni, platform director at SciLifeLab, notes, "AF_unmasked is a user-friendly tool that benefits the cryo-EM community by analyzing large, flexible molecular assemblies."
The project began as a knowledge-building effort to understand new prediction tools but quickly evolved beyond the state-of-the-art, demonstrating the collaborative power of SciLifeLab’s experimental and computational expertise.
AF_unmasked will soon be available through SciLifeLab Serve. Stay updated by visiting our Serve website in the coming days.
