Protein Prediction Shrinks from Months to Hours, MassiveFold Outperforms AlphaFold3

MassiveFold significantly improves protein structure prediction efficiency through optimization and parallelization, reducing computation time from months to hours and supporting multiple hardware settings; combined with extended sampling and template-less dropout methods, MassiveFold outperforms AlphaFold3 in CASP15 target testing, generating more diverse structures with high confidence; MassiveFold is scalable to different research environments, solving the bottleneck of AlphaFold in terms of GPU resource requirements and operation efficiency, and advancing the progress of protein modeling.