Igor Baskin - Exhaustive QSPR studies of a large diverse set of ionic liquids: How accurately can we predict melting points?

Document created by Igor Baskin on Aug 22, 2014
Version 1Show Document
  • View in full screen mode

  Publication Details (including relevant citation   information):

  Varnek, A.; Kireeva, N.; Tetko, I. V.; Baskin, I. I.; Solov'ev,   V. P. Exhaustive QSPR studies of a large diverse set of ionic   liquids: How accurately can we predict melting points? J.   Chem. Inf. Mod. 2007, 47 (3), 1111-1122.

  Abstract:

  Several popular machine learning methods-Associative Neural   Networks (ANN), Support Vector Machines (SVM), k Nearest   Neighbors (kNN), modified version of the partial least-squares   analysis (PLSM), backpropagation neural network (BPNN), and   Multiple Linear Regression Analysis (MLR)-implemented in ISIDA,   NASAWIN, and VCCLAB software have been used to perform QSPR   modeling of melting point of structurally diverse data set of 717   bromides of nitrogen-containing organic cations (FULL) including   126 pyridinium bromides (PYR), 384 imidazolium and   benzoimidazolium bromides (IMZ), and 207 quaternary ammonium   bromides (QUAT). Several types of descriptors were tested:   E-state indices, counts of atoms determined for E-state atom   types, molecular descriptors generated by the DRAGON program, and   different types of substructural molecular fragments. Predictive   ability of the models was analyzed using a 5-fold external   cross-validation procedure in which every compound in the parent   set was included in one of five test sets. Among the 16 types of   developed structure - melting point models, nonlinear SVM, ASNN,   and BPNN techniques demonstrate slightly better performance over   other methods. For the full set, the accuracy of predictions does   not significantly change as a function of the type of   descriptors. For other sets, the performance of descriptors   varies as a function of method and data set used. The root-mean   squared error (RMSE) of prediction calculated on independent test   sets is in the range of 37.5-46.4 ?C (FULL), 26.2-34.8 ?C (PYR),   38.8-45.9 ?C (IMZ), and 34.2-49.3 ?C (QUAT). The moderate   accuracy of predictions can be related to the quality of the   experimental data used for obtaining the models as well as to   difficulties to take into account the structural features of   ionic liquids in the solid state (polymorphic effects, eutectics,   glass formation).

  Address (URL): http://pubs.acs.org/doi/abs/10.1021/ci600493x

 

Attachments

    Outcomes