A systematic modeling methodology of deep neural network-based structure-property relationship for rapid and reliable prediction on flashpoints

Deep neural networks (DNNs) based quantitative structure-property relationship (QSPR) studies are receiving increasing attention due to their excellent performances. A systematic methodology coupling multiple machine learning technologies is proposed to systematically solve vital problems including applicability domain and prediction uncertainty in DNN-based QSPR modeling. 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 prediction 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. Furthermore, the proposed methodology breaks through difficulties in analyzing the uncertainty of DNN-based QSPRs and presents an AD correlated with the uncertainty.

Automated summary

The research proposes a systematic approach to solving key problems in DNN-based QSPR modeling, including applicability domain and prediction uncertainty, using multiple machine learning technologies. The method extracts features through principal component analysis and kernel PCA, defines a detailed applicability domain using the K-means algorithm, and further analyzes prediction uncertainty.