In this work quantitative structure-retention relationship models were developed to predict the solute retention factors in supercritical fluid chromatography (SFC) in various organic modifiers. Data set contains the retention data of 35 various organic compounds in 0, 2, 4 and 6% of methanol in mobile phase. The obtained 140 data points were divided into training, internal and external test sets which have 93, 23 and 24 retention data. The diversity validation test was performed on data the set to ensure that the structure of the training and/or test sets can represent those of the whole ones. Descriptors which were selected by stepwise multiple linear regression (MLR) methods are: the percent of organic modifier in mobile phase, salvation connectivity index chi-2, salvation connectivity index chi-5, H attached to heteroatom, the 2(nd) structural information content and polarity parameter. These descriptors were used as features in generation of linear and non-linear models using MLR and artificial neural network (ANN) methods, respectively. The root mean square error of MLR model are 0.116, 0.138 and 0.260 for training, internal and external test sets, respectively, while these values are 0.036, 0.097 and 0.244 for ANN model, respectively. Comparison between these values and other statistical parameters obtained from these two models reveals the credibility of ANN in prediction of solute retention factors in SFC.

Multiple linear regression (MLR) and artificial neural network (ANN) were used to predict the migration factors of benzene derivatives in MEKC. Some topological and electronic descriptors were calculated for each solute in the data set, and then the stepwise MLR method was used to select more significant descriptors and MLR model development. The selected descriptors are: Kier & Hall index (order1), relative negative charge surface area, HA dependent HDSA-2/TMSA, C component of moment of inertia, Y component of dipole moment and SDS to decanol ratio in mobile phase. In the next step these descriptors were used as input of an ANN. After optimization and training of ANN it was used to predict the migration factors of external test set as well as internal and training sets. The root mean square errors for ANN predicted migration factors of training, internal and external test set were 0.110, 0.231 and 0.228, respectively, while these values are 0.200, 0.240 and 0.247 for the MLR model, respectively. Comparison between these values and other statistical parameters for these two models revealed that there were not any significant differences between ANN and MLR in prediction of solute migration factors in MEKC.

In this work multiple linear regression (MLR) was carried out for the prediction of immobilized artificial membrane (IAM) retention factors of 40 basic and neutral drugs in two mobile phase compositions. We developed some MLR models by using linear free energy relationships (LFER) parameters and also theoretically derived molecular descriptor. Root mean square error of MLR model in prediction of log k(wPBS)(IAM) and k(wMOPS)(IAM) are 0.332 and 0.351, respectively, while these values are 0.371 and 0.500 for LFER models. Inspections to these values indicate that the statistical parameters of MLR models are better than LFER models. The credibility of MLR models was evaluated by using leave-many-out cross-validation and y-scrambling procedures. The results of these tests indicate the applicability of theoretically derived molecular descriptors and LFER parameters prediction of IAM retention of drugs.