4D-Fingerprint Categorical QSAR Models for Skin Sensitization Based on the Classification of Local Lymph Node Assay Measures

Abstract

Currently, the only validated methods to identify skin sensitization effects are in vivo models, such as the Local Lymph Node Assay (LLNA) and guinea pig studies. There is a tremendous need, in particular due to novel legislation, to develop animal alternatives, eg. Quantitative Structure-Activity Relationship (QSAR) models. Here, QSAR models for skin sensitization using LLNA data have been constructed. The descriptors used to generate these models are derived from the 4D-molecular similarity paradigm and are referred to as universal 4D-fingerprints. A training set of 132 structurally diverse compounds and a test set of 15 structurally diverse compounds were used in this study. The statistical methodologies used to build the models are logistic regression (LR), and partial least square coupled logistic regression (PLS-LR), which prove to be effective tools for studying skin sensitization measures expressed in the two categorical terms of sensitizer and non-sensitizer. QSAR models with low values of the Hosmer-Lemeshow goodness-of-fit statistic, , are significant and predictive. For the training set, the cross-validated prediction accuracy of the logistic regression models ranges from 77.3% to 78.0%, while that of PLS-logistic regression models ranges from 87.1% to 89.4%. For the test set, the prediction accuracy of logistic regression models ranges from 80.0%-86.7%, while that of PLS-logistic regression models ranges from 73.3%-80.0%. The QSAR models are made up of 4D-fingerprints related to aromatic atoms, hydrogen bond acceptors and negatively partially charged atoms.

A Minireview of Available Skin Sensitization (Q)SARs/Expert Systems

“…The feasibility of using 4D fingerprints derived from a methodology called 4D molecular fingerprint similarity analysis was investigated to develop QSAR models for skin sensitization [71]. An early version of the LLNA dataset [62] comprising data on 219 chemicals was sorted into three categories; non-sensitizers (NS) and weak sensitizers (WS), (101 chemicals); moderate sensitizers (MS; 72 chemicals), and strong (SS) or extreme (ES) sensitizers (46 chemicals).…”

Categorical QSAR models for skin sensitization based on local lymph node assay measures and both ground and excited state 4D-fingerprint descriptors

“…The predicted probabilities, classifications and accuracies for models (14), (15) and (16), based on the training set compounds, are shown in Table 1. A summary of classification accuracy for the test set predictions is given in Table 2.…”

“…Presumably certain ground state descriptors provide glimpses of the reactivity behavior of a molecule. For example, we have carried out a two-state categorical QSAR modeling of skin sensitization [13] using a dataset constructed from the validated in vivo murine local lymph node assay (LLNA) [14]. In our initial study, ground state 4D-fingerprints (4D-FP) were used as a descriptor set to generate categorical QSAR models for two states: sensitizer and nonsensitizer [13].…”

“…For example, we have carried out a two-state categorical QSAR modeling of skin sensitization [13] using a dataset constructed from the validated in vivo murine local lymph node assay (LLNA) [14]. In our initial study, ground state 4D-fingerprints (4D-FP) were used as a descriptor set to generate categorical QSAR models for two states: sensitizer and nonsensitizer [13]. This current study focused on exploring the usefulness of the statistical methodologies of logistic regression (LR) and partial-least square coupled logistic regression (PLS-CLR) to build two-state categorical models.…”

“…In a more recent study, we used the same LLNA dataset and both LR and PLS-CLR to build 3-state and two-2-state categorical models for skin sensitization potency [15]. The three-state QSAR model yields a classification accuracy of 73.4% for the training set and 63.6% for the test set, while the random average value of classification accuracy for any 3-state dataset is 33.3%.…”