Predicting pK_a by Molecular Tree Structured Fingerprints and PLS

Abstract

This is the second phase of the pK(a) predictor published earlier (J. Chem. Inf. Comput. Sci. 2002, 42, 796-805). The algorithm has been extended by treating specific chemical classes separately and generating tree-structured molecular descriptors tailored to each individual class. A training set consisting of 625 acids and 412 bases covers the major areas of chemical space involved in protonation and deprotonation. The models obtained demonstrate excellent statistics (SE = 0.41 for acids and 0.30 for bases) and yielded accurate predictions on an external test set. The quality and statistical performance of pK(a) prediction has been improved considerably over the initial implementation of the method.

Predicting and Tuning Physicochemical Properties in Lead Optimization: Amine Basicities

“…In a second phase, Xing and co-workers refined their algorithm for selected and well-represented subclasses of basic and acidic compounds, resulting in improved predictions, albeit at the cost of uniform applicability. [42] Eckert and Klamt used an algorithm based on improved dielectric continuum solvation methods (DCSMs), also accounting for short-range electrostatics of polar solutes and ions as well as hydrogen bonding. [43] Compared with other algorithms, the model was calibrated with a relatively small dataset of 43 bases, excluding aliphatic amines.…”

Modern methods for estimation of ionization constants of organic compounds in solution

“…This procedure has become very popular for prediction of many other physicochemical parameters of compounds and their biological activity [150,[176][177][178][179][180][181][182][183][184][185][186][187][188]. Extensive use of QSPR in solving various problems of physical organic chemistry is favored by development of computational methods and instruments and statistical processing procedures such as principal component analysis (PCA) [154,[163][164][165]185], independent component analysis (ICA) [189], partial least squares (PLS) [153,154,156,157,168,174] in combination with iterative variable elimination (IVE-PLS) [156], multilinear regression (MLR) [155,157,159,160,175,187], and variable importance in projection (VIP) [190,191].…”

“…Examples of molecular-tree structured fingerprint (MTSF) of morpholine molecule as numerical notation of the dependence of minimal energies (E min ) of interaction between O-probe and morpholine molecule on the topological distance (TP) to the oxygen atom therein [167]. QSPR calculations of pK a 's also utilized such descriptors as charges on atoms [37,38,158,175,190,201], group philicity [197], parameters of atoms relevant to molecular mechanics (SYBYL-type atoms) [173,174], empirical parameters included into free online SPARC package (see below) [198], semiempirical descriptors [37, 38, 163-165, 171, 172, 175, 190], and (less frequently) experimental physicochemical parameters [151,[163][164][165]. In the past decade, topological descriptors were used most extensively, in particular fragment descriptors [187], topological distances [167,173,174], 3D molecular interaction fields [167], molecular-tree structured fingerprints (MTSF) [162,167,173,174], and "quantum chemical topology" descriptors calculated in terms of the quantum similarity theory (QST) [161,169].…”

“…In the past decade, topological descriptors were used most extensively, in particular fragment descriptors [187], topological distances [167,173,174], 3D molecular interaction fields [167], molecular-tree structured fingerprints (MTSF) [162,167,173,174], and "quantum chemical topology" descriptors calculated in terms of the quantum similarity theory (QST) [161,169]. The most effective descriptor is the sum by numerical parameters of types of atoms and bonds (SYBYL) at a distance of up to 5 bonds from the reaction center [167,173,174]. An example of using such approach is shown in Fig.…”

“…The above listed descriptors are obtained by processing of the results of semiempirical calculations for large arrays of compounds and functional fragments by topological methods, comparative molecular field analysis (CoMFA) [167,173,174] and comparative molecular surface analysis (CoMSA) [153,154,156,168]. The latter implies calculation of 3D energy profile for interaction of a molecule with a probe atom (SYBYL atom) which is virtually placed in different positions nearly molecular skeleton [156,167] (Fig.…”