Zhang Fan, Hirama Yui, Onishi Shintaro, Mori Takuya, Ono Naoaki, Kanaya Shigehiko
Material Science Research, Kao Corporation, 1334 Minato, Wakayama-shi 640-8580, Wakayama, Japan.
Division of Information Science, Graduate School of Science and Technology, Nara Institute of Science and Technology, 8916-5 Takayama-cho, Ikoma 630-0192, Nara, Japan.
Microorganisms. 2024 Jul 31;12(8):1568. doi: 10.3390/microorganisms12081568.
In case of future viral threats, including the proposed Disease X that has been discussed since the emergence of the COVID-19 pandemic in March 2020, our research has focused on the development of antiviral strategies using fragrance compounds with known antiviral activity. Despite the recognized antiviral properties of mixtures of certain fragrance compounds, there has been a lack of a systematic approach to optimize these mixtures. Confronted with the significant combinatorial challenge and the complexity of the compound formulation space, we employed Bayesian optimization, guided by Gaussian Process Regression (GPR), to systematically explore and identify formulations with demonstrable antiviral efficacy. This approach required the transformation of the characteristics of formulations into quantifiable feature values using molecular descriptors, subsequently modeling these data to predict and propose formulations with likely antiviral efficacy enhancements. The predicted formulations underwent experimental testing, resulting in the identification of combinations capable of inactivating 99.99% of viruses, including a notably efficacious formulation of five distinct fragrance types. This model demonstrates high predictive accuracy (coefficient determination Rcv2 > 0.7) and suggests a new frontier in antiviral strategy development. Our findings indicate the powerful potential of computational modeling to surpass human analytical capabilities in the pursuit of complex, fragrance-based antiviral formulations.
在未来面临病毒威胁时，包括自2020年3月新冠疫情爆发以来一直被讨论的拟议中的X疾病，我们的研究聚焦于利用具有已知抗病毒活性的香料化合物开发抗病毒策略。尽管某些香料化合物混合物具有公认的抗病毒特性，但一直缺乏优化这些混合物的系统方法。面对重大的组合挑战和化合物配方空间的复杂性，我们采用了以高斯过程回归（GPR）为指导的贝叶斯优化方法，系统地探索和识别具有可证明抗病毒功效的配方。这种方法需要使用分子描述符将配方特征转化为可量化的特征值，随后对这些数据进行建模，以预测和提出可能具有增强抗病毒功效的配方。对预测的配方进行了实验测试，结果确定了能够灭活99.99%病毒的组合，包括一种由五种不同香料类型组成的特别有效的配方。该模型显示出较高的预测准确性（决定系数Rcv2>0.7），并为抗病毒策略开发开辟了新的前沿领域。我们的研究结果表明，在追求基于香料的复杂抗病毒配方方面，计算建模具有超越人类分析能力的强大潜力。
