Enzymes play an important role in industry due to their catalytic properties and environmental friendliness. For application in harsh industrial environments, enzymes are modified to obtain improved stability through simultaneous mutations at multiple sites. Contrary to experimental methods, computational methods are significantly more efficient and cost-effective for screening stabilizing mutations. However, there is no systematic evaluation of multi-point mutation predictors for enzyme applications. In this study, we investigate computational methods that are published for multi-point mutation stability change and select four representative ones: DDGun, MAESTRO, DynaMut2 and DDMut. To evaluate their performance comprehensively, we collect three benchmark datasets with different numbers of mutation sites and three enzyme-only datasets, utilizing RMSE and PCC as metrics. Then, we compare their ability to identify stabilizing (0) and destabilizing (G¡0) variants. Meanwhile, we analyze the robustness of four predictors by testing their prediction biases on stabilizing and destabilizing variants. The results of the above tasks indicate that DDMut displays exceptional efficiency among all predictors. Furthermore, to explore the availability of four predictors on enzyme modification, we gather 167 IsPETase variants and 34 leaf-branch compost cutinase (LCC) variants from the literature. DDGun achieves the highest Recall on both single-point and multi-point mutations of them. We also evaluate the four predictors from the perspective of amino acid hydrophobicity and charge change at the mutation site. Overall, DDGun is the optimal choice for enzyme thermostability modification and DDMut can also be taken into consideration. We hope our work can guide enzyme engineering in various mutation scenarios.

中文翻译：

---

利用计算工具快速筛选酶热稳定性修饰的多点突变


酶由于其催化特性和环境友好性而在工业中发挥着重要作用。为了在恶劣的工业环境中应用，酶经过修饰，通过多个位点的同时突变获得更高的稳定性。与实验方法相反，计算方法对于筛选稳定突变明显更有效且更具成本效益。然而，目前还没有对酶应用的多点突变预测因子进行系统评估。在本研究中，我们研究了已发表的多点突变稳定性变化的计算方法，并选择了四种代表性的方法：DDGun、MAESTRO、DynaMut2 和 DDMu。为了全面评估它们的性能，我们收集了三个具有不同数量突变位点的基准数据集和三个仅酶数据集，利用 RMSE 和 PCC 作为指标。然后，我们比较它们识别稳定 (0) 和不稳定 (G¡0) 变体的能力。同时，我们通过测试四个预测变量对稳定和不稳定变体的预测偏差来分析它们的稳健性。上述任务的结果表明 DDMu 在所有预测器中表现出卓越的效率。此外，为了探索酶修饰的四个预测因子的可用性，我们从文献中收集了 167 个 IsPETase 变体和 34 个叶枝堆肥角质酶 (LCC) 变体。 DDGun 在单点突变和多点突变上均实现了最高的 Recall。我们还从氨基酸疏水性和突变位点电荷变化的角度评估了四个预测因子。总体而言，DDGun是酶热稳定性修饰的最佳选择，DDMu也可以考虑。 我们希望我们的工作能够指导各种突变情况下的酶工程。