6 Mode-of-Action-Guided, Molecular …
105
and maintained by the National Center for Biotechnology Information (NCBI), now
contains 144,042 records of resolved structures (as of September 7, 2018) for chemically bound or unbound proteins, DNAs and RNAs (as well as their complexes).
Consequently, recent years have seen increased applications of molecular docking
in qualitative toxicological MoA studies; for example: (1) screening of endocrine
disrupting environmental compounds through docking to the ligand-binding domain
of estrogen receptor α [43], (2) predicting idiosyncratic drug reactions via examining
the binding modes of drugs in the human leukocyte antigens [44], and (3) evaluating
the endocrine disrupting activity of 45 bisphenol A (BPA) replacement compounds
using molecular dynamics (MD) simulations [45].
6.2 Methodology
6.2.1 Approach Overview
In view of the existing limitations of current approaches briefly reviewed above,
we have developed a novel, state-of-the-art predictive toxicology approach that is
guided by MoA, MIE or other toxicological mechanism information at the molecular level, as illustrated in Fig. 6.1. This approach is also based on molecular modeling that integrates structural biology principles, computational chemistry tools, and
machine learning techniques. Historically, molecular modeling (molecular docking
and MD simulation in particular) has been applied to qualitative studies for elucidating the mechanism of molecular interactions between a ligand and a target biomacromolecule [45, 46] or to high-throughput preliminary screening of drug candidates for
their potency on disease targets [34, 35, 47]. Here we expand its application to quantitative toxicity assessment. Meanwhile, machine learning (especially deep learning)
methods that have been widely applied in predictive toxicology are employed to train
and validate toxicity prediction models for qualitative categorization and quantitative
estimation of uncharacterized chemicals [48, 49].
6.2.2 Approach Implementation
Our approach is implemented through a target-specific toxicity knowledgebase
(TsTKb) that consists of a pre-categorized database of chemical MoAs (ChemMoA)
and a library of pre-built, category-specific classification and quantification models
(see [50] for more information). ChemMoA serves as the depository of chemicals
with known MoAs or MIEs (i.e., known target biomacromolecules) and quantitative
information for measured toxicity endpoints (Fig. 6.1). The following information is
curated in ChemMoA: chemical data (e.g., IUPAC name, identifier, SMILES structure, and 1D to 3D molecular descriptors), target data (i.e., the 3D structures of
105
and maintained by the National Center for Biotechnology Information (NCBI), now
contains 144,042 records of resolved structures (as of September 7, 2018) for chemically bound or unbound proteins, DNAs and RNAs (as well as their complexes).
Consequently, recent years have seen increased applications of molecular docking
in qualitative toxicological MoA studies; for example: (1) screening of endocrine
disrupting environmental compounds through docking to the ligand-binding domain
of estrogen receptor α [43], (2) predicting idiosyncratic drug reactions via examining
the binding modes of drugs in the human leukocyte antigens [44], and (3) evaluating
the endocrine disrupting activity of 45 bisphenol A (BPA) replacement compounds
using molecular dynamics (MD) simulations [45].
6.2 Methodology
6.2.1 Approach Overview
In view of the existing limitations of current approaches briefly reviewed above,
we have developed a novel, state-of-the-art predictive toxicology approach that is
guided by MoA, MIE or other toxicological mechanism information at the molecular level, as illustrated in Fig. 6.1. This approach is also based on molecular modeling that integrates structural biology principles, computational chemistry tools, and
machine learning techniques. Historically, molecular modeling (molecular docking
and MD simulation in particular) has been applied to qualitative studies for elucidating the mechanism of molecular interactions between a ligand and a target biomacromolecule [45, 46] or to high-throughput preliminary screening of drug candidates for
their potency on disease targets [34, 35, 47]. Here we expand its application to quantitative toxicity assessment. Meanwhile, machine learning (especially deep learning)
methods that have been widely applied in predictive toxicology are employed to train
and validate toxicity prediction models for qualitative categorization and quantitative
estimation of uncharacterized chemicals [48, 49].
6.2.2 Approach Implementation
Our approach is implemented through a target-specific toxicity knowledgebase
(TsTKb) that consists of a pre-categorized database of chemical MoAs (ChemMoA)
and a library of pre-built, category-specific classification and quantification models
(see [50] for more information). ChemMoA serves as the depository of chemicals
with known MoAs or MIEs (i.e., known target biomacromolecules) and quantitative
information for measured toxicity endpoints (Fig. 6.1). The following information is
curated in ChemMoA: chemical data (e.g., IUPAC name, identifier, SMILES structure, and 1D to 3D molecular descriptors), target data (i.e., the 3D structures of