Addressing the Metabolic Stability of Antituberculars through Machine Learning

Article abstract of DOI:10.1021/acsmedchemlett.7b00299

We present the first prospective application of our mouse liver microsomal (MLM) stability Bayesian model. CD117, an antitubercular thienopyrimidine tool compound that suffers from metabolic instability (MLM t_1/2 < 1 min), was utilized to assess the predictive power of our new MLM stability model. The S-substituent was removed, a set of commercial reagents was utilized to construct a virtual library of 411 analogues, and our MLM stability model was applied to prioritize 13 analogues for synthesis and biological profiling. In MLM stability assays, all 13 analogues had superior metabolic stability to the parent compound, and six new analogues had acceptable MLM t_1/2 values greater than or equal to 60 min. It is noteworthy that whole-cell efficacy and lack of relative mammalian cell cytotoxicity could not be predicted simultaneously. These results support the utility of our new MLM stability model in chemical tool and drug discovery optimization efforts.

Full text of DOI:10.1021/acsmedchemlett.7b00299

Subscriber access provided by University of Sydney Library
Letter
Addressing the Metabolic Stability of
Antituberculars through Machine Learning
Thomas P. Stratton, Alexander Luke Perryman, Catherine Vilchèze, Riccardo Russo, Shao-Gang Li,
Jimmy S Patel, Eric Singleton, Sean Ekins, Nancy Connell, William R Jacobs, and Joel S. Freundlich
ACS Med. Chem. Lett., Just Accepted Manuscript • DOI: 10.1021/acsmedchemlett.7b00299 • Publication Date (Web): 14 Sep 2017
Downloaded from http://pubs.acs.org on September 21, 2017
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
ACS Medicinal Chemistry Letters is published by the American Chemical Society. 1155
Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.

Page 1 of 7
ACS Medicinal Chemistry Letters
1
2
3
4
5
6
7
8
9
Addressing the Metabolic Stability of Antituberculars through Ma-
chine Learning
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Thomas P. Stratton,1,†,‡ Alexander L. Perryman,1,‡ Catherine Vilchèze,2 Riccardo Russo,3 ShaoꢀGang
Li,1 Jimmy S. Patel,1 Eric Singleton,3 Sean Ekins,4,5 Nancy Connell,3 William R. Jacobs Jr.,2 and Joel S.
Freundlich1,3,*
1. Department of Pharmacology, Physiology, and Neuroscience, Rutgers University – New Jersey Medical School, Newark,
New Jersey, 07103, USA.
2. Howard Hughes Medical Institute, Department of Microbiology and Immunology, Albert Einstein College of Medicine,
Bronx, NY 10461, USA.
3. Division of Infectious Disease, Department of Medicine and the Ruy V. Lourenço Center for the Study of Emerging and
Reꢀemerging Pathogens, Rutgers University ꢀ New Jersey Medical School, Newark, New Jersey 07103, USA.
4. Collaborative Drug Discovery, 1633 Bayshore Highway, Suite 342, Burlingame, CA 94010, USA.
5. Collaborations Pharmaceuticals, Inc., 840 Main Campus Drive, Lab 3510, Raleigh, NC 27606, USA.
KEYWORDS: Bayesian, mouse liver microsomal stability, antitubercular, chemical tool optimization, computer-aided ana-
log design, machine learning
We present the first prospective application of our mouse liver microsomal (MLM) stability Bayesian model. CD117, an antituberꢀ
cular thienopyrimidine tool compound that suffers from metabolic instability (MLM t1/2 < 1 min), was utilized to assess the predicꢀ
tive power of our new MLM stability model. The Sꢀsubstituent was removed, a set of commercial reagents was utilized to construct
a virtual library of 411 analogs, and our MLM stability model was applied to prioritize 13 analogs for synthesis and biological proꢀ
filing. In MLM stability assays, all 13 analogs had superior metabolic stability to the parent compound, and 6 new analogs had acꢀ
ceptable MLM t1/2 values greater than or equal to 60 min. It is noteworthy that wholeꢀcell efficacy and lack of relative mammalian
cell cytotoxicity could not be predicted simultaneously. These results support the utility of our new MLM stability model in chemiꢀ
cal tool and drug discovery optimization efforts.
Mycobacterium tuberculosis, the pathogen that causes tuꢀ
berculosis, is responsible for a global health pandemic. It kills
approximately 1.4 million people each year.1 Multiꢀdrug reꢀ
sistant, extensivelyꢀdrug resistant, and even totallyꢀdrug reꢀ
sistant strains of M. tuberculosis continue to increase in freꢀ
quency and in global distribution. Consequently, new drugs to
treat M. tuberculosis infections via novel therapeutic apꢀ
proaches represent an urgent public health necessity. To inꢀ
crease the efficiency and accuracy of discovering novel chemꢀ
ical tools that can (a) facilitate studies of the fundamental biꢀ
ology of M. tuberculosis and (b) lay the foundation for the
discovery and development of new drugs to treat drugꢀ
resistant infections, we are creating, honing, and applying
novel computational tools and workflows.2 Our approaches,
and many of the specific machine learning models that we
develop, could also be applied to help advance translational
research against other diseases. We are (1) developing various
types of machine learning models and (2) combining them
with each other and with other computational techniques, folꢀ
lowed by (3) prospective predictions and (4) experimental
validation, to help overcome each of the hurdles that a comꢀ
pound must pass before it can become either a useful chemical
tool for M. tuberculosis studies or an antitubercular drug lead.
In the search for small molecule agents that act as chemical
tools to probe basic biology as well as to seed novel therapeuꢀ
tic strategies, we must consider as early as possible key moꢀ
lecular properties. In our experience, liver microsomal stabilꢀ
ity, kinetic solubility, in vitro efficacy and mammalian cell
cytotoxicity are amongst the most important metrics to be
profiled.3ꢀ5 Given our focus on small molecules as antitubercuꢀ
lar agents,6, 7 we are interested in assessing mouse liver microꢀ
somal (MLM) stability. Although mouse models fail to recaꢀ
pitulate the entire disease pathology observed for M. tubercu-
losis infection of humans (e.g., necrotic lesions such as caseꢀ
ous granulomas are not formed in most types of mice),8, 9 the
field has witnessed a significant correlation between acceptaꢀ
ble efficacy outcomes in humans and in mice.10, 11 Thus, we
assert that given the goal of demonstrating a small molecule’s
ability to significantly reduce M. tuberculosis infection in
mice, we should initially assess the MLM stability of a chemiꢀ
cal tool or drug discovery hit.
The typical MLM assay involves determination of the staꢀ
bility of a small molecule in the presence of a microsomal
preparation, which contains the enzymes involved in phase I
(oxidative) metabolism, except for some MLM assays that
ACS Paragon Plus Environment

ACS Medicinal Chemistry Letters
Page 2 of 7
contain both phase I and phase II (conjugative) metabolism.12
rated version of publicly deposited MLM data sets to identify
1
2
3
4
5
6
7
8
The half life (t1/2) and intrinsic clearance (Clint) are determined.
In our programs, we have set the goal for acceptable MLM
stability as t1/2 ≥ 60 min and Clint ≤ 10 µL/min/mg protein.13
Once a small molecule’s MLM stability has been assessed, it
may also be helpful to determine what metabolites are formed
via mass spectroscopic techniques, followed by synthesizing
the dominant metabolites and then assaying them for inhibiꢀ
tion of M. tuberculosis growth.7 We herein characterize the in
vitro wholeꢀcell growth inhibition of the laboratory wild type
M. tuberculosis strain H37Rv with an MIC, or minimum inꢀ
hibitory concentration equal to the concentration of compound
capable of inhibiting 90% of bacterial growth as judged by a
microplate Alamar blue assay.14
what molecular physiochemical and structural features are
associated with sufficient or insufficient MLM stability, as
judged by our metric of stable compounds having a t1/2 ≥ 60
min. Using a novel approach to the construction of machine
learning models, we discovered that pruning the training set
(by discarding moderately stable / moderately unstable models
from the training set, such that sufficiently stable molecules
were defined by t1/2 ≥ 60 min and insufficiently stable moleꢀ
cules were set as those with a t1/2 < 30 min, while the intermeꢀ
diate compounds were deleted) led to models that were better
able to predict their respective training set as well as indeꢀ
pendent external test and validation sets (when compared to
the full or unꢀpruned model). In particular, we previously conꢀ
cluded the pruned t1/2 MLM Bayesian model constructed with
nine descriptors was most useful for filtering a large library of
compounds to harvest candidates with MLM stability. Conꢀ
versely, we observed that the full t1/2 MLM Bayesian model
constructed using only the FCFPꢀ6 descriptor (Full
MLM/FCFPꢀ6 model which had a full training set but a
pruned set of descriptors) performed the best when identifying
a small number of candidates with MLM stability.13 We herein
describe the first prospective application of the Full t1/2 MLM
Bayesian model constructed using only the FCFPꢀ6 descriptor
to enhance the MLM stability of CD117.
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
We have utilized such early profiling techniques to assess
the strengths and shortcomings of a class of antitubercular
agents stemming from thienopyrimidine CD117 (Figure 1).15
CD117 displays promising wholeꢀcell efficacy versus in vitro
cultured M. tuberculosis (MIC = 0.19 – 0.38 µg/mL) and seꢀ
lectivity versus Vero cells as a model mammalian cell line (SI
= CC50/MIC = 120 – 240, where CC50 = minimum concentraꢀ
tion to inhibit the growth of 50% of the Vero cells). However,
CD117 suffers from rapid metabolism, in the presence of
MLM (t1/2 < 1 min) or when dosed to BALB/c mice, to generꢀ
ate the corresponding wholeꢀcell inactive carboxylic acid meꢀ
tabolite (JSFꢀ2000). While this decomposition does not occur
within the bacterial growth medium, metabolomics studies
confirmed this degradation does take place to a minor extent
within CD117ꢀtreated M. tuberculosis over 18 h. Given that
the primary target of CD117 was (and still remains) a subject
of intense study, we have been unable to inform our optimizaꢀ
tion efforts with details as to what protein/s CD117 is/are
binding and how its ethyl ester is interacting with the protein/s
at an atomic level. This is not an uncommon circumstance
with antibacterial chemical biology and drug discovery reꢀ
search and in particular within the tuberculosis field.16 Thus,
we turned to medicinal chemistry heuristics and attempts to
mimic structural aspects of the ethyl ester of CD117 with a
heterocyclic isostere led to isoxazole (JSFꢀ2070) and thiazole
(JSFꢀ2088) analogs with MLM t1/2 values of 19.0 and 66.6
min, respectively.7
Figure 2. Computational workflows focused on enhancing the
MLM stability of CD117.
Our approach began with the obvious decision that the metꢀ
abolically labile ethyl ester moiety had to be replaced. Leverꢀ
aging our published route to CD117 and its analogs (Scheme
1), we chose to consider a library of 411 alkyl bromides and
aldehydes (which could be converted into alkyl bromides via
straightforward reduction and bromination) available from
SigmaꢀAldrich. A virtual library was created and then scored
via three different workflows (Figure 2; Supporting Inforꢀ
mation for an Excel file with all 411 candidates scored with all
models). The majority of the candidates occupy different
property space from those CD117 analogs previously syntheꢀ
sized (Supporting Information Figure S1). In workflow (i), the
library was scored with the Full MLM/FCFPꢀ6 model and the
top 20 scoring molecules were noted.
Figure 1. Structure of CD117 with an emphasis on its metabolic
liability, and the previous medicinal chemistry optimization to
improve metabolic stability that afforded JSFꢀ2070 and JSFꢀ2088.
While pursuing further evolution of these heterocyclic reꢀ
placements for the ethyl ester, we sought to design and then
apply an orthogonal strategy that would more broadly consider
chemical space and not be limited by our medicinal chemistry
insights, which are undoubtedly biased by our collective expeꢀ
rience. We turned to our recently devised naïve Bayesian
(heretofore referred as Bayesian) classifier models for MLM
stability.13 The models have essentially learned from our cuꢀ
Next, we sought to predict molecules with sufficient MLM
stability as well as acceptable antitubercular wholeꢀcell efficaꢀ
cy and relative Vero cell cytotoxicity. The intersection was
probed for virtual library members predicted to be MLM staꢀ
ble through the Full MLM/FCFPꢀ6 model and active (acceptaꢀ
ACS Paragon Plus Environment

Page 3 of 7
ACS Medicinal Chemistry Letters
ble MIC versus M. tuberculosis)/nonꢀcytotoxic (acceptable
tures, nitro aromatics were not triaged due to their prevalence
in antiꢀinfectives.20 We do note that the potential exists for
other liabilities to have been introduced into the candidates as
the models were only trained to learn about MLM t1/2 or
{MLM t1/2, MIC, and SI}.
1
2
3
4
5
6
7
8
CC50 versus Vero cell cells as judged through the SI value)
with our published TAACFꢀCB2 dualꢀevent Bayesian model
(defining an active such that MIC ≤ 10 µg/mL and SI ≥ 10),
which was trained with a 1,783 molecule library subset as
reported by the Southern Research Institute.17 However, none
of the 411 candidates was predicted to be MLM stable, wholeꢀ
cell active, and sufficiently nonꢀcytotoxic. To refine the dualꢀ
event model we pruned its training set (defining an active such
that MIC ≤ 10 µg/mL and SI ≥ 15 instead of SI ≥ 10) to genꢀ
erate the CB2v2_SI15 dualꢀevent Bayesian (further model
details are in the Supporting Information). Alteration of the
training set active definition allowed the model to consider
different areas of chemical space as being more favorable to
metabolic stability. We also explored the use of both FCFP12
and FCFP6 descriptors to investigate whether these new
CB2v2_SI15 models would (a) enhance the accuracy of rankꢀ
ing the 88 previously synthesized CD117 analogs,7 as an exꢀ
ternal test of the accuracy of these new models (see Table S2
in the Supporting Information) given our specific focus on
thienopyrimidines in this optimization, and (b) produce new
consensus workflows that allowed some analogs in the new
virtual library to pass through these filters (Figure 2, workflow
ii). Both of these goals were achieved with the new FCFP12ꢀ
based CB2v2_SI15 model. This new model was found to exꢀ
hibit acceptable (ROC ≥ 0.7)13 internal and external statistics
of 0.73 and 0.76, respectively (Tables S1 and S2 in the Supꢀ
porting Information). In workflow (ii), the consensus scoring
of the virtual library of analogs using this revised model and
the Full MLM/FCFPꢀ6 model afforded ten candidates predictꢀ
ed to meet the criteria of MLM stable, wholeꢀcell active, and
relatively nonꢀcytotoxic.
Scheme 1. Synthesis of CD117 Analogsa
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
aReagents and conditions: (a) NCCH2CN, S8,
Et2NH/EtOH; (b) PhC(O)NCS, acetone; (c) NaOH(aq),
EtOH, microwave, (d) RCH2Br.
The synthetic efforts recalled our previously published
methodology,7 based on the two step/one pot cyclizationꢀ
alkylation of the key benzoyl thiourea intermediate 1 (Scheme
1). Compared to our previous route, we were pleased to find
that microwave heating significantly decreased the time for the
baseꢀpromoted cyclization to the proposed thiolate intermediꢀ
ate. The electrophiles chosen to react with this intermediate to
form the resulting sulfide were commercially available or preꢀ
pared by straightforward functional group interconversions
(reduction followed by bromination) from the commercially
available aldehyde. The resulting CD117 analogs were characꢀ
terized by NMR spectroscopy (1H and 13C), HPLC, and
HRMS (Supporting Information).
Each of the synthesized candidates was assayed for MLM
stability, as judged primarily by t1/2 given the training criterion
of the Bayesian models. All thirteen analogs displayed larger
MLM t1/2 values than the parent compound CD117 (Table 1).
Six of the thirteen analogs displayed MLM t1/2 values above
our stability criterion of 60 min, which represents a hit rate of
46% (6/13). One analog displayed a t1/2 of 58.2 min, which is
very close to our intended goal and could also be viewed as
exhibiting sufficient MLM stability. Only three of the thirteen
analogs (23%) had an MLM t1/2 less than 30 min. Interestingly,
the MLM stability model fared similarly, with or without the
inclusion of predictions from the dualꢀevent model for wholeꢀ
cell efficacy and lack of relative Vero cell cytotoxicity. Failing
to predict a metabolically stable compound with an extremely
limited test set (N = 1), workflow (iii) and the potential to
leverage a chemical series Bayesian model will need to be
investigated with a significantly larger virtual library of anaꢀ
logs. While the MLM models were not trained with Clint data,
one may also employ a stability criterion of MLM Clint ≤ 10
µL/min/mg protein. Through this lens, three of the thirteen
candidates (23%) may be judged suitably stable, with another
five analogs (38%) nearly meeting this criterion with 10 < Clint
< 15.
In addition, the concept of a chemical series specific Bayesꢀ
ian model for wholeꢀcell efficacy was explored. We utilized
the
Collaborative
Drug
Discovery
software
(www.collaborativedrug.com) to create a singleꢀevent model,
using 88 previously synthesized CD117 analogs7 and the acꢀ
tivity definition of MIC ≤ 10 µg/mL (further model details are
in the Supporting Information). With acceptable internal statisꢀ
tics (see Figure S5 in the Supporting Information), the
CDD_CD117 model was used in workflow (iii) to score the
virtual library. The intersection of those compounds predicted
to be active with the CDD_CD117 model and stable with the
Full MLM/FCFPꢀ6 model consisted of four molecules.
Finally, visual inspection of the total candidates from the
three workflows was conducted to remove molecules with
reactive functionality (e.g., an electrophile such as an aldeꢀ
hyde, epoxide, or alkyl halide),18 a molecular weight in excess
of 600 g/mol, or an AlogP > 6.0.19 The resulting thirteen canꢀ
didates for synthesis and biological testing (Table 1 excluding
CD117 and INH controls) were comprised of six from the Full
MLM/FCFPꢀ6 model, six from this model in consensus with
the CB2v2_SI15 dual event Bayesian, and one in consensus
with the CDD_CD117 model. Although these candidates were
selected using the Full MLM/FCFPꢀ6 model, they were all
also predicted to be stable with the pruned t1/2 MLM Bayesian
model constructed using all nine descriptors. While these thirꢀ
teen candidates cluster more closely to the previously syntheꢀ
sized CD117 analogs than the initial set of 411 (Supporting
Information Figures S1 and S2), their respective sulfide penꢀ
dant moieties contain different chemotypes than those previꢀ
ously explored through a medicinal chemistry heuristic apꢀ
proach.7 As evident from inspection of the candidate strucꢀ
Table 1. Biological profiling of computationally evolved
analogs of CD117
g
Entry
Structure
MLM
Clearꢀ
MICf CC50
45
Stabilityd ancee
<1
>1390 0.19
CD117
–
ACS Paragon Plus Environment

ACS Medicinal Chemistry Letters
Page 4 of 7
0.38
124 candidate analogs (reduced from the list of 411 comꢀ
pounds via the previously mentioned molecular weight and
1
2
3
4
5
6
7
8
120
5.80
12.5
– 25
>50
50
JSFꢀ2450a
JSFꢀ2509a
JSFꢀ2486a
JSFꢀ2489b
JSFꢀ2790b
AlogP filters), prepared, and assayed for MLM t1/2 (see Table
S4 in the Supporting Information). All 5 analogs in this ranꢀ
dom set were unstable, and the Bayesian model’s prediction
was correct for only one 1 of 5 candidates. This is consistent
with the proposed workflow and our previous studies showing
that the Full MLM/FCFPꢀ6 model performed best when utiꢀ
lized to select the topꢀ or bottomꢀscoring candidates.13 Given
the fundamentals of Bayesian modeling, one has the greatest
confidence in the highestꢀscoring entities as their probability
of stability scales with their Bayesian score. Inversely, the
lowest scoring candidates have the least likelihood of being
stable.
95.0
87.7
7.30
7.90
25
50
–
>200 12.5
>200 >50
64.2
60.8
10.8
11.4
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
10
25
100
6.2
60.3
11.5
The novel workflows presented herein represent a compleꢀ
mentary approach to the utilization of medicinal chemistry
heuristics when faced with the challenge of enhancing the
metabolic stability of a chemical tool and/or drug discovery
entity. We have demonstrated the utility of a Bayesian model,
cognizant of the physiochemical and structural features related
to metabolic stability in the presence of mouse liver microꢀ
somes. Regardless of the inclusion of a dualꢀevent Bayesian
model to account for wholeꢀcell activity and lack of relative
Vero cell cytotoxicity, our chosen MLM Bayesian model corꢀ
rectly selected for novel CD117 analogs with acceptable MLM
stability 50% of the time, while triaging 80% of the unstable
(and lowest scoring) candidates.
JSFꢀ2812b
58.2
47.5
11.9
14.6
200
10
50
JSFꢀ2449a
JSFꢀ2452a
>200
JSFꢀ2433a
JSFꢀ2771b
34.3
31.8
21.7
20.1
21.8
31.9
>200 >50
>200 12.5
100 – 1.6
200
JSFꢀ2432c
This outcome represents a significant extension from our
previous work,13 which made available to the scientific comꢀ
munity machine learning models for MLM stability while
pharmaceutical industry models have been published but not
disclosed due to intellectual property concerns.21 We are unaꢀ
ware of other published reports where such a model has been
used alone, or in combination with efficacy and cytotoxicity
models, in prospective predictions to guide compound optimiꢀ
zation. This machine learning approach is complementary to
MetaSite,22 which computationally predicts the metabolic hot
spots of small molecules due to phase I metabolism.
JSFꢀ2772b
JSFꢀ2430b
19.7
17.0
35.1
40.8
200
3.1
>200 25
N.D.
N.D.
0.039 N.D.
INH
a-cCorrespond to workflow used as described in Figure 2 repreꢀ
d
senting (i), (ii), and (iii), respectively. MLM stability from assay
as represented by t1/2 (min). eMLM Clint as represented by
f
g
µL/min/mg protein. MIC against Mtb H37Rv in µg/mL. Vero
Further work is required to understand how in vitro efficacy,
Vero cell cytotoxicity, and MLM stability can be predicted
simultaneously, at the risk of certain features intrinsic to each
model not being shared (i.e., one structural feature or physioꢀ
chemical property may correlate strongly with MLM stability
and yet be inconsistent or inversely correlated with wholeꢀcell
activity and/or Vero cell cytotoxicity). In fact, medicinal
chemistry optimizations involve even more criteria beyond
those considered in this report,23 and ultimately we will search
for a subset of those that can be successfully predicted simulꢀ
taneously. The evolved CD117 analogs disclosed here are now
the subjects of downstream pharmacokinetic studies to enable
consideration for in vivo assays of antitubercular efficacy.
Coupled with insights from our ongoing mechanistic studies of
these thienopyrimidines, we aim for nearꢀterm in vivo validaꢀ
tion of their novel mechanism of action.
Cell toxicity as represented by CC50 in µg/mL.
The candidates were also assessed for their antitubercular
wholeꢀcell efficacy and Vero cell cytotoxicity. The most poꢀ
tent antitubercular compounds were JSFꢀ2452 and 2790, exꢀ
hibiting an MIC value of 10 µg/mL. The Vero cell cytotoxiciꢀ
ties of these two compounds were >200 µg/mL (SI > 20) and
100 µg/mL (SI = 10), respectively. The consensus approach in
workflow (ii) delivered one analog (JSFꢀ2790) with sufficient
MLM t1/2 and MIC, although its SI value of 10 fell short of the
target value of 15. Thus, at this juncture, workflow (ii) does
not add significant value over workflow (i). Workflow (i)
lacked inclusion of a dualꢀevent model but did produce one
compound (JSFꢀ2452) with acceptable MLM t1/2, wholeꢀcell
activity and relative Vero cell cytotoxicity.
To further demonstrate the predictive power of our Bayesian
approach, we synthesized the lowest scoring 5 CD117 candiꢀ
dates according to the Full MLM/FCFPꢀ6 model. The model
correctly predicted the metabolic instability of 4/5 compounds
(see Table S3 in the Supporting Information), demonstrating
its ability to triage analogs unlikely to exhibit a sufficient t1/2.
Using the random number generator functionality within Liꢀ
breOffice Calc (www.libreoffice.org), 5 candidate analogs
(not previously assessed) were chosen randomly from a set of
ASSOCIATED CONTENT
Supporting Information
The Supporting Information is available free of charge on the
ACS Publications website at DOI: .
ACS Paragon Plus Environment

Page 5 of 7
ACS Medicinal Chemistry Letters
5. Ekins, S.; Freundlich, J. S.; Hobrath, J. V.; Lucile White, E.; Reynꢀ
Details with regard to the computational workflows and inherent
olds, R. C. Combining computational methods for hit to lead optimiꢀ
zation in Mycobacterium tuberculosis drug discovery. Pharm. Res.
2014, 31, 414ꢀ35.
6. Ekins, S.; Perryman, A. L.; Clark, A. M.; Reynolds, R. C.; Freunꢀ
dlich, J. S. Machine Learning Model Analysis and Data Visualization
with Small Molecules Tested in a Mouse Model of Mycobacterium
tuberculosis Infection (2014ꢀ2015). J. Chem. Inf. Model. 2016, 56,
1332ꢀ43.
7. Li, S.ꢀG.; Vilcheze, C.; Chakraborty, S.; Wang, X.; Kim, H.; Aniꢀ
setti, M.; Ekins, S.; Rhee, K. Y.; Jacobs Jr., W. R.; Freundlich, J. S.
Evolution of a thienopyrimidine antitubercular relying on medicinal
chemistry and metabolomics insights. Tetrahedron Lett. 2015, 56,
3246ꢀ3250.
8. Kramnik, I.; Beamer, G. Mouse models of human TB pathology:
roles in the analysis of necrosis and the development of hostꢀdirected
therapies. Semin. Immunopathol. 2016, 38, 221ꢀ37.
9. Harper, J.; Skerry, C.; Davis, S. L.; Tasneen, R.; Weir, M.;
Kramnik, I.; Bishai, W. R.; Pomper, M. G.; Nuermberger, E. L.; Jain,
S. K. Mouse model of necrotic tuberculosis granulomas develops
hypoxic lesions. J. Infect. Dis. 2012, 205, 595ꢀ602.
10. Lanoix, J. P.; Lenaerts, A. J.; Nuermberger, E. L. Heterogeneous
disease progression and treatment response in a C3HeB/FeJ mouse
model of tuberculosis. Dis. Model. Mech. 2015, 8, 603ꢀ10.
11. Irwin, S. M.; Driver, E.; Lyon, E.; Schrupp, C.; Ryan, G.; Gonzaꢀ
lezꢀJuarrero, M.; Basaraba, R. J.; Nuermberger, E. L.; Lenaerts, A. J.
Presence of multiple lesion types with vastly different microenvironꢀ
ments in C3HeB/FeJ mice following aerosol infection with Mycobacꢀ
terium tuberculosis. Dis. Model. Mech. 2015, 8, 591ꢀ602.
12. Di, L.; Kerns, E. H.; Ma, X. J.; Huang, Y.; Carter, G. T. Applicaꢀ
tions of high throughput microsomal stability assay in drug discovery.
Comb. Chem. High Throughput Screen. 2008, 11, 469ꢀ76.
13. Perryman, A. L.; Stratton, T. P.; Ekins, S.; Freundlich, J. S. Preꢀ
dicting Mouse Liver Microsomal Stability with "Pruned" Machine
Learning Models and Public Data. Pharm. Res. 2016, 33, 433ꢀ49.
14. Palomino, J. C.; Martin, A.; Camacho, M.; Guerra, H.; Swings, J.;
Portaels, F. Resazurin microtiter assay plate: simple and inexpensive
method for detection of drug resistance in Mycobacterium tuberculoꢀ
sis. Antimicrob. Agents Chemother. 2002, 46, 2720ꢀ2.
15. Vilcheze, C.; Baughn, A. D.; Tufariello, J.; Leung, L. W.; Kuo,
M.; Basler, C. F.; Alland, D.; Sacchettini, J. C.; Freundlich, J. S.;
Jacobs, W. R., Jr. Novel inhibitors of InhA efficiently kill Mycobacteꢀ
rium tuberculosis under aerobic and anaerobic conditions. Antimicrob.
Agents Chemother. 2011, 55, 3889ꢀ98.
16. Cooper, C. B. Development of Mycobacterium tuberculosis whole
cell screening hits as potential antituberculosis agents. J. Med. Chem.
2013, 56, 7755ꢀ60.
17. Ekins, S.; Reynolds, R. C.; Kim, H.; Koo, M.ꢀS.; Ekonomidis, M.;
Talaue, M.; Paget, S. D.; Woolhiser, L. K.; Lenaerts, A.; Bunin, B.
A.; Connell, N.; Freundlich, J. S. Bayesian models leveraging bioacꢀ
tivity and cytotoxicity information for drug discovery. Chem. Biol.
2013, 20, 370ꢀ378.
18. Walters, W. P.; Stahl, M. T.; Murcko, M. A. Virtual screening ꢀ an
overview. Drug. Discov. Today 1998, 3, 160ꢀ178.
19. Ghose, A. K.; Viswanadhan, V. N.; Wendoloski, J. J. Prediction
of Hydrophobic (Lipophilic) Properties of Small Organic Molecules
Using Fragmental Methods: An Analysis of ALOGP and CLOGP
Methods. J. Phys. Chem. A. 1998, 102, 3762ꢀ3772.
20. Thompson, A. M.; O'Connor, P. D.; Marshall, A. J.; Yardley, V.;
Maes, L.; Gupta, S.; Launay, D.; Braillard, S.; Chatelain, E.;
Franzblau, S. G.; Wan, B.; Wang, Y.; Ma, Z.; Cooper, C. B.; Denny,
W. A. 7ꢀSubstituted 2ꢀNitroꢀ5,6ꢀdihydroimidazo[2,1ꢀb][1,3]oxazines:
Novel Antitubercular Agents Lead to a New Preclinical Candidate for
Visceral Leishmaniasis. J Med Chem 2017, 60, 4212ꢀ4233.
21. Hu, Y.; Unwalla, R.; Denny, R. A.; Bikker, J.; Di, L.; Humblet, C.
Development of QSAR models for microsomal stability: identificaꢀ
tion of good and bad structural features for rat, human and mouse
microsomal stability. J. Comput. Aided Mol. Des. 2010, 24, 23ꢀ35.
22. Cruciani, G.; Carosati, E.; De Boeck, B.; Ethirajulu, K.; Mackie,
C.; Howe, T.; Vianello, R. MetaSite: understanding metabolism in
1
2
3
4
5
6
7
8
Bayesian models, syntheses and characterization data for the reꢀ
ported compounds, and protocols for the biological assays.
The Supporting Information is available free of charge on the
ACS Publications website.
Supplemental file with spreadsheet of computational predictions
for all 411 candidates (.xlsx)
9
AUTHOR INFORMATION
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Corresponding Author
*Phone: +1ꢀ973ꢀ972ꢀ7165. Eꢀmail: freundjs@rutgers.edu.
Present Addresses
†Current address: Department of Chemistry, The Scripps Research
Institute, La Jolla, California 92037, USA.
Author Contributions
T.P.S., S.G.L., and J.S.P. synthesized the compounds. A.L.P.
constructed and validated the Bayesian models and utilized them
to score the virtual library of candidates. C.V. conducted the M.
tuberculosis growth inhibition assays. R.R. and E.S. carried out
the Vero cell cytotoxicity assays. J.S.F., S.E., W.R.J., and N.C.
directed the research. J.S.F. conceived of and originated the proꢀ
ject. J.S.F., A.L.P., T.P.S., and S.E. wrote the manuscript. All
authors contributed to editing the manuscript. All authors have
given approval to the final version of the manuscript. ‡These auꢀ
thors contributed equally.
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENT
J.S.F. and S.E. acknowledge support from award number
R44TR000942ꢀ02 “Biocomputation across distributed private
datasets to enhance drug discovery” from the National Institutes
of Health and National Center for Advancing Translational Sciꢀ
ences. J.S.F. and N.C. acknowledge support from award number
1U19AI109713 NIH/NIAID for the “Center to develop therapeuꢀ
tic countermeasures to highꢀthreat bacterial agents,” from the
National Institutes of Health: Centers of Excellence for Translaꢀ
tional Research (CETR). W.R.J. acknowledges support from the
National Institutes of Health awards AI026170 and
U19AI111276. J.S.F. and S.E. acknowledge BIOVIA for kindly
providing Discovery Studio and Pipeline Pilot.
ABBREVIATIONS
MLM, mouse liver microsomal; NMR, nuclear magnetic resoꢀ
nance; HPLC, highꢀperformance liquid chromatography; HRMS,
highꢀresolution mass spectrometry.
REFERENCES
1. Global Tuberculosis Report; WHO: Geneva, 2016.
2. Ekins, S.; Freundlich, J. S.; Choi, I.; Sarker, M.; Talcott, C. Comꢀ
putational databases, pathway and cheminformatics tools for tubercuꢀ
losis drug discovery. Trends Microbiol. 2011, 19, 65ꢀ74.
3. Workman, P.; Collins, I. Probing the probes: fitness factors for
small molecule tools. Chem. Biol. 2010, 17, 561ꢀ77.
4. Lakshminarayana, S. B.; Huat, T. B.; Ho, P. C.; Manjunatha, U. H.;
Dartois, V.; Dick, T.; Rao, S. P. Comprehensive physicochemical,
pharmacokinetic and activity profiling of antiꢀTB agents. J Antimi-
crob. Chemother. 2015, 70, 857ꢀ67.
ACS Paragon Plus Environment

ACS Medicinal Chemistry Letters
Page 6 of 7
human cytochromes from the perspective of the chemist. J. Med.
Chem. 2005, 48, 6970ꢀ9.
1
23. Wermuth, C. G.; Aldous, D.; Raboisson, P.; Rognan, D. Eds. The
Practice of Medicinal Chemistry; Elsevier: Amsterdam, 2015.
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

Page 7 of 7
ACS Medicinal Chemistry Letters
Insert Table of Contents artwork here
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
7
ACS Paragon Plus Environment