No document available.
Abstract :
[en] QSRR models were developed for retention prediction of
small molecules using data at five pH conditions
(2.0, 3.5, 5.0, 6.5, and 8.0) and for two gradient times of
20 and 60 minutes. Molecular descriptors were derived
from the chemical structures of the compounds using
SMILES and proportions of each form of the compound of
interest at the corresponding pH. MLR, Boruta, Lasso,SR
were used as feature selection methods. One unique
feature set was generted based on mutually inclusive
features among base feature selection algorithms. 60
mixed qsrr models were developed using algorithms -
LR,SVR,RF,GBM and stacked ensemble learning. The best
model was then used to predict the retention times of the
external test compound at each pH condition. We
propose that our strategy involving stacked ensemble
learning is very unique and effective method for retention
prediction of small molecules.
