A systematic DNN-based QSPR modeling methodology for rapid and reliable prediction on flashpoints of chemicals

Abstract

Quantitative structure-property relationship (QSPR) studies based on deep neural networks (DNN) are receiving increasing attention due to their excellent performances. A systematic methodology coupling multiple machine learning technologies is proposed to solve vital problems including applicability domain and prediction uncertainty in DNN-based QSPRs. Key features are rapidly extracted from plentiful but chaotic descriptors by principal component analysis (PCA) and kernel PCA. Then, a detailed applicability domain (AD) is defined by K-means algorithm to avoid unreliable predictions and discover its potential impact on uncertainty. Moreover, prediction uncertainty is analyzed with dropout-embedded DNN by thousands of independent tests to assess the reliability of predictions. The prediction of flashpoint temperature is employed as a case study demonstrating that the model accuracy is remarkably improved comparing with the referenced model. More importantly, the proposed methodology breaks through difficulties in analyzing the uncertainty of DNN-based QSPRs and presents an AD correlated with the uncertainty.

Predicting structure‐dependent properties directly from the three dimensional molecular images via convolutional neural networks

“…points are quite precise. Nevertheless, the accuracy on the estimation of flash points and critical points is not as good as other method such as group contribution and QSPR, 22,41 mainly because the CNN method requires a large dataset for training, which is not available in this work. In a word, the 3D graph provides an alternative representation for molecules, and can serve as a simple but generalized method for property prediction, in the field of both quantum calculated and experimental properties.…”

“…In chemical computations such DFT and molecular dynamics, the system energy is the basic property, and can be further used in We further trained the model with experimental boiling points of more than 12,000 species consisting of 10 elements (Table 1), and the results is displayed in Figure 7A1 We also used this method to predict the flash point, an important property to characterize the flammability of organic components under heat. 41 The available samples for flash points are only about cup, but we could not distinguish which method was employed from database, which may also contribute to the final prediction error. different physical properties may have a similar dependence on structural features.…”

“…We also used this method to predict the flash point, an important property to characterize the flammability of organic components under heat 41 . The available samples for flash points are only about 1/3 of that for boiling or melting points.…”