Published: CABEQ 36 (4) (2022) 223-230
Paper type: Original Scientific Paper
This work applies the concept of structural causal modelling (SCM) for the prediction of eutectic temperatures of choline chloride based deep eutectic solvents (DES). Two SCM models were developed, one based on molecular descriptors (MD), and the other based on molecular fingerprints (MF). The models are presented in the form of directed acyclic graphs (DAG). The SCM-MD model shows that the chi simple cluster connectivity descriptor (SC.5) and a number of hydrogen atoms (nH.1) are the key causal variables. The causal relations between the model variables and eutectic temperature were determined after performing d-separation to block the variable confounding interference. The corresponding nonlinear causal relations were modelled by Bayes neural network with a single inner layer. Based on the SCM-MD model, a decision tree is proposed for the prediction of eutectic temperatures. Model performances were tested on a literature dataset of eutectic temperatures of ChCl based DESs. The SCM-MD model provided the most accurate prediction with an error of 7.5 °C.
This work is licensed under a Creative Commons Attribution 4.0 International License
DES, eutectic temperature, causal AI, molecular descriptors, molecular fingerprints