discovery. Databases such as NCBI Epigenomics, HEMD, ChEpimod, ChEMBL,
and ChromoHub provide access only to epigenetic proteins and their phylogenetic
associations. However, their chemical libraries are mainly based on virtual inhibitors of epigenetic proteins (e.g., PubChem bioassay data) [58]. To retain only
validated molecular scaffolds that actually interact with epigenetic targets, we
collected data manually from the literature for experimentally verified inhibitors.
We designed EpiDBase, to view, explore, search, and analyze the small-molecule
modulators targeting various epigenetic protein families. EpiDBase is manually
curated to identify unique molecular scaffolds and includes text search, structural
editor, and chemical fingerprint search, for powerful browsing [46]. Molecules from
EpiDBase can be selected and further derivatized to obtain a potent lead toward an
epigenetic protein in drug discovery.
Strategy 2. Weed out the problematic compounds
One of the key aspects in the prioritization of chemical matter is to weed out the
‘problematic’ or ‘risky’ compounds. Such compounds called ‘PAINS’ are ‘frequent
hitters’ and belong to a subset of chemical substructures that interact
non-specifically with proteins in various unrelated bioassays. PAINS can readout as
false positives due to non-selective binding with proteins [59], fluorescence [60],
redox activity [61], cysteine oxidation [62], and aggregation [63]. More than 450
structural classes [64] have been identified as PAINS, and a typical academic
screening library may consist of 5–12% of such compounds [65]. The most
recurring chemotypes are rhodanines, phenol-sulfonamides, toxoflavins, isothiazolones, enones, curcumin, hydroxyphenylhydrazones, quinones, and catechol. For
example, rhodanines are reported as promising bioactive compounds, but they may
undergo light-induced reactions and modify some proteins covalently [66].
Likewise, phenol-sulfonamides are unstable compounds and can alter the redox
cycle and covalently modify the target proteins. Unfortunately, several articles and
patents in the literature include PAINS as potential bioactive compounds [67–70].
A medicinal chemists‘ precise look at the structure can help the biologist to exclude
such compounds. Recently, chemical substructure filters and rules (e.g., PAINS,
REOS, and others [49, 71–74]) have been introduced to identify these problematic
moieties. In EpiDBase, Eli Lilly MedChem regular rules were applied to weed out
such promiscuous compounds [46]. The Eli Lilly MedChem rules are defined by a
set of 275 rules and are capable of identifying compounds that may interfere with
biological assays. We processed a total of 5401 molecules, out of which 1664
molecules were rejected by the filter rules. The remaining 3737 molecules can be
further exploited in epigenetic drug discovery and can prove promising.
Strategy 3. Perform the computational ADME/toxicity prediction
Majority of drug candidates fail in clinical trials owing to their unfavorable
absorption, distribution, metabolism, excretion, and toxicity (ADMET) profile
(Fig. 3) [75–77]. A recent study [2] showed that only 32% of Phase II drug
Integrated Chemoinformatics Approaches …
257
and ChromoHub provide access only to epigenetic proteins and their phylogenetic
associations. However, their chemical libraries are mainly based on virtual inhibitors of epigenetic proteins (e.g., PubChem bioassay data) [58]. To retain only
validated molecular scaffolds that actually interact with epigenetic targets, we
collected data manually from the literature for experimentally verified inhibitors.
We designed EpiDBase, to view, explore, search, and analyze the small-molecule
modulators targeting various epigenetic protein families. EpiDBase is manually
curated to identify unique molecular scaffolds and includes text search, structural
editor, and chemical fingerprint search, for powerful browsing [46]. Molecules from
EpiDBase can be selected and further derivatized to obtain a potent lead toward an
epigenetic protein in drug discovery.
Strategy 2. Weed out the problematic compounds
One of the key aspects in the prioritization of chemical matter is to weed out the
‘problematic’ or ‘risky’ compounds. Such compounds called ‘PAINS’ are ‘frequent
hitters’ and belong to a subset of chemical substructures that interact
non-specifically with proteins in various unrelated bioassays. PAINS can readout as
false positives due to non-selective binding with proteins [59], fluorescence [60],
redox activity [61], cysteine oxidation [62], and aggregation [63]. More than 450
structural classes [64] have been identified as PAINS, and a typical academic
screening library may consist of 5–12% of such compounds [65]. The most
recurring chemotypes are rhodanines, phenol-sulfonamides, toxoflavins, isothiazolones, enones, curcumin, hydroxyphenylhydrazones, quinones, and catechol. For
example, rhodanines are reported as promising bioactive compounds, but they may
undergo light-induced reactions and modify some proteins covalently [66].
Likewise, phenol-sulfonamides are unstable compounds and can alter the redox
cycle and covalently modify the target proteins. Unfortunately, several articles and
patents in the literature include PAINS as potential bioactive compounds [67–70].
A medicinal chemists‘ precise look at the structure can help the biologist to exclude
such compounds. Recently, chemical substructure filters and rules (e.g., PAINS,
REOS, and others [49, 71–74]) have been introduced to identify these problematic
moieties. In EpiDBase, Eli Lilly MedChem regular rules were applied to weed out
such promiscuous compounds [46]. The Eli Lilly MedChem rules are defined by a
set of 275 rules and are capable of identifying compounds that may interfere with
biological assays. We processed a total of 5401 molecules, out of which 1664
molecules were rejected by the filter rules. The remaining 3737 molecules can be
further exploited in epigenetic drug discovery and can prove promising.
Strategy 3. Perform the computational ADME/toxicity prediction
Majority of drug candidates fail in clinical trials owing to their unfavorable
absorption, distribution, metabolism, excretion, and toxicity (ADMET) profile
(Fig. 3) [75–77]. A recent study [2] showed that only 32% of Phase II drug
Integrated Chemoinformatics Approaches …
257