Exploring the nematic transition temperatures of pyridine-containing liquid crystalline compounds by explainable machine learning model

By the combination application of machine learning methods and SHapley Additive exPlanations (SHAP) analysis, explainable machine learning models were built for the first time to explore the nematic–isotropic transition behaviour of pyridine-containing liquid crystalline compounds. All models were fully validated. The results show that these models have high goodness of fit, robustness and prediction ability. Among them, the random forest (RF) performs best, exhibiting the highest values of Q² (0.964), QF12(0.962),QF22(0.962), QF32(0.966) and CCC (0.980) and the smallest errors (MAE = 9.516K, RMSE = 13.899K and AARD% = 2.203) for the test set. The RF model was visualised and explained using the SHapley Additive exPlanations (SHAP) method, shedding light on the ‘black-box’ nature of the machine learning model, and thus enhancing its transparency and credibility. A Williams plot was used to characterise the applicability domain range of the RF model and tree manifold approximation and projection (TMAP) technology was utilised to illustrate the similarity and diversity of the entire data space to assist in the analysis of outliers. In addition, the odd-even effect is discussed. The research in this work will aid in the in silico QSPR prediction of nematic–isotropic transition temperatures for compounds not synthesised or tested and can also be extended to other studies.