Ras GTP酶在细胞信号转导途径中起着关键作用。Ras基因的突变在约三分之一的癌细胞系中发生，并且通常与不良的临床预后相关。热点残基Gly12、Gly13和Gln61涵盖了97%的致癌突变，这些突变损害了Ras的酶活性。使用QM/MM自由能计算，我们提出了一种双步机制来描述由野生型Ras.GAP复合物催化的GTP水解过程。我们发现，通过Gln61作为瞬态Brønsted碱来发生催化性水的去质子化。我们还使用QM/MM最小化确定了关键致癌Ras突变体G12D和G12C的反应轨迹，与实验观察到的催化活性丧失相吻合，从而验证了我们的反应机制。利用优化后的反应路径，我们设计了一种快速准确的过程来设计激活G12D Ras的GAP突变体。我们替换了活性位点附近的GAP残基，并确定了190个单突变体的激活能垒。我们还构建了一个用于超快筛选的机器学习模型，通过快速预测能垒高度，在单一突变和双突变上进行了测试。这项研究表明，通过QM/MM反应路径优化可以实现快速准确的筛选，以设计具有增强催化活性的蛋白质序列。预测将出现几个GAP突变，可以重新启用致癌G12D的催化活性，为克服异常的Ras驱动信号转导提供了有前途的途径，而不是抑制酶活性。所概述的计算筛选方案同样适用于设计类似的配体和辅基。

Ras GTPases play a crucial role in cell signaling pathways. Mutations of the Ras gene occur in about one third of cancerous cell lines and are often associated with detrimental clinical prognosis. Hot spot residues Gly12, Gly13, and Gln61 cover 97% of oncogenic mutations, which impair the enzymatic activity in Ras. Using QM/MM free energy calculations, we present a two-step mechanism for the GTP hydrolysis catalyzed by the wild-type Ras.GAP complex. We found that the deprotonation of the catalytic water takes place via the Gln61 as a transient Brønsted base. We also determined the reaction profiles for key oncogenic Ras mutants G12D and G12C using QM/MM minimizations, matching the experimentally observed loss of catalytic activity, thereby validating our reaction mechanism. Using the optimized reaction paths, we devised a fast and accurate procedure to design GAP mutants that activate G12D Ras. We replaced GAP residues near the active site and determined the activation barrier for 190 single mutants. We furthermore built a machine learning for ultrafast screening, by fast prediction of the barrier heights, tested both on the single and double mutations. This work demonstrates that fast and accurate screening can be accomplished via QM/MM reaction path optimizations to design protein sequences with increased catalytic activity. Several GAP mutations are predicted to re-enable catalysis in oncogenic G12D, offering a promising avenue to overcome aberrant Ras-driven signal transduction by activating enzymatic activity instead of inhibition. The outlined computational screening protocol is readily applicable for designing ligands and cofactors analogously.