xTrimoABFold: Antibody Structure Prediction Model and xTrimoDock: Complex Structure Prediction Model
In October 2022, BioMap's research team introduced antibody-related prediction models under its xTrimo system, recognized as a global leader in life science models. Two models - 'xTrimoAB Fold' and 'xTrimoDock', have shown substantial advancement in the field of antibody structure prediction.
The xTrimo system is a biologically multimodal pre-trained model system capable of characterizing monomeric proteins, protein interactions, immune cells, systems and beyond, offering vital computational biology tools for immune drug discovery.
Among these, xTrimoDock enables simultaneous utilization of both chain sequence and structure data to precisely forecast multi-chain protein structures. Its prediction precision surpassed competing methods by over 80%, attaining remarkable 1-2 angstroms accuracy for antibody-antigen complex predictions, similarly accurate to cryo-electron microscopy resolutions.
Even further, this model requires just half an hour, drastically reducing the processing time for complex structure prediction.
In comparison, xTrimoAB Fold outperforms MSA-based AlphaFold2 by nearly 151 times, in terms of antibody structure prediction speed, surpassing all recent SOTA algorithms like OmegaFold, HelixFold-Single, and IgFold, with increased RMSD performance exceeding 30%.
On this achievement, CTO at BioMap, Song Le, noted, "From predicting protein structures to identifying interactions between proteins, BioMap aims to decode the 'language' of life and develop groundbreaking pharmaceuticals to improve human health through the large-scale xTrimo pre-trained model."