ABSTRACT
Topological Structure-Information Representation (SIR) serves as the basis for QSAR model development on two data sets of dipeptides. Data sets of both bitter-taste (48 compounds) and angiotensin-converting-enzyme (ACE) inhibition (58 compounds) were analyzed by means of multiple linear-regression methods to produce QSAR models that relate structure to property. For the bitter-taste data set, two variables describe the data well, both being whole-molecule descriptors: (1)chi(v) (molecular connectivity first-order valence index) and SHBa (sum of E-State indices for H-bond acceptors) yield r(2)=0.88, s=0.22. External validation and cross-validation indicate that the model may be predictive. For the ACE-inhibition data set, five variables produced a satisfactory model. Four of the descriptors relate to amino acid side chains: the E-State polarity/non-polarity index Q(v) (for position A adjacent to the N-terminus; Fig. 1) and the E-State index s(2) (for the backbone position of substitution), along with the square of the molecular connectivity path-four valence index ((4)chi(PC); for side chain B adjacent to C-terminus) and the E-State index s(5) (for the attachment point of the side chain B (Fig. 1)). Together with the E-State whole-molecule descriptor for internal H-bonding (five skeletal bonds; SHBint5), the five variables form a predictive model (r(2)=0.88, s=0.36). Both external-test and cross-validation-test statistics indicate that the model may be predictive. This study is the first investigation in which E-State descriptors are developed for amino acid side chains.
