期刊
JOURNAL OF CHEMICAL INFORMATION AND MODELING
卷 62, 期 22, 页码 5397-5410出版社
AMER CHEMICAL SOC
DOI: 10.1021/acs.jcim.2c00841
关键词
-
类别
资金
- Defense Advanced Research Projects Agency [HR00111920025]
- Combat Capabilities Development Command (DEVCOM) Army Research Laboratory [W15P7T-19-D-0126]
This study demonstrates the application of transfer learning in predicting experimentally measured properties of molecules. By training a regression model on a small number of molecules with measured values and a large number of molecules with computed properties, the researchers achieved higher prediction accuracy compared to direct machine learning and physics-based models. The findings show that the characteristics of the computed dataset and the architecture of the model play significant roles in improving prediction accuracy for small experimental datasets.
For many experimentally measured chemical properties that cannot be directly computed from first-principles, the existing physics-based models do not extrapolate well to out-of-sample molecules, and experimental datasets themselves are too small for traditional machine learning (ML) approaches. To overcome these limitations, we apply a transfer learning approach, whereby we simultaneously train a multi-target regression model on a small number of molecules with experimentally measured values and a large number of molecules with related computed properties. We demonstrate this methodology on predicting the experimentally measured impact sensitivity of energetic crystals, finding that both characteristics of the computed dataset and model architecture are important to prediction accuracy of the small experimental dataset. Our directed-message passing neural network (D-MPNN) ML model using transfer learning outperforms direct-ML and physics-based models on a diverse test set, and the new methods described here are widely applicable to modeling many other structure-property relationships.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据