Nature | 从核磁共振谱的缺失信息中解析蛋白质毫秒尺度动力学

TL;DR - A Nature study introduces Dyna-1, which learns millisecond-scale protein dynamics from missing NMR signals. It turns previously discarded spectral gaps into weak supervision for predicting functionally important conformational changes.

• Researchers built RelaxDB from 133 proteins and mBMRB from 9,381 proteins with missing backbone ^15N assignments.
• Dyna-1 combines ESM-3 sequence and structure representations, achieving about 0.74 AUROC under strict homology-separated validation.
• Independent relaxation and CPMG experiments showed that it predicts real µs–ms exchange, including dynamics missed by conventional CPMG analysis.
• Prospective experiments confirmed broad exchange in Chitinase 19 and minimal exchange in yjbJ, broadly matching model predictions.