Content uploaded by Peter Fedichev
Author content
All content in this area was uploaded by Peter Fedichev on Mar 18, 2020
Content may be subject to copyright.
AI for the repurposing of approved or investigational drugs against COVID-19
Konstantin Avchaciov, Olga Burmistrova, and Peter O. Fedichev∗
GERO PTE. LTD., 60 Paya Lebar Road #05-40B Paya Lebar Square, Singapore 409051
The coronavirus disease 2019 (COVID-19) caused by severe acute respiratory syndrome coron-
avirus 2 (SARS-CoV-2), has spread globally since 2019 and reached the pandemic level in 2020.
We utilized a deep neural network to search for host-target acting antivirals among experimental
and approved drugs with potential activity against coronavirus-born diseases. To achieve the goal
we searched for gene expression signatures of molecular perturbations most closely resembling the
effects of the COBP2 gene knockout since COBP2 is required for replication of a genetically similar
virus SARS-CoV. The majority of the top-scoring molecules were already suggested for repurposing
as broad-spectrum antivirals. One of the approved drugs from the list, nitazoxanide, has recently
demonstrated activity against SARS-CoV-2. We, therefore, urge prompt experimental characteriza-
tion of the other predictions and highlight the potential of modern AI/ML technologies for prompt
identification of human-trial ready therapeutics against the world’s most urgent medical needs. We
encourage academic and industrial collaborations to validate the results of this research and further
develop the most successful compounds.
INTRODUCTION
On December 30, 2019, according to some sources,
an artificial-intelligence (AI) company called BlueDot
alerted clients to an unusual bump in pneumonia cases in
Wuhan, China. It was nine days before the World Health
Organization (WHO) officially flagged what we have all
come to know as Coronavirus disease 2019 (COVID-19).
This is an infectious disease caused by severe acute res-
piratory syndrome coronavirus 2 (SARS-CoV-2, previ-
ously referred to as 2019-nCoV). The disease has spread
globally since 2019, resulting in the 2019–20 coronavirus
pandemic. While the majority of cases result in mild
symptoms, some progress to pneumonia and multi-organ
failure [1].The death rate per the number of diagnosed
cases is estimated at between 1% and 5% and varies by
confounding health conditions and availability of health-
care resources [2].
The application of AI technologies is not limited to
sophisticated natural language processing for outbreak
monitoring. In just a few recent weeks we saw announce-
ments from InSilico Medicine and DeepMind joining the
global effort against the disease by contributing their
AI/ML pipelines for the research and drug discovery com-
munity. The companies respectively produced and made
public domain a set of putative small molecules inhibitors
of one of the key virus targets, 3C-like Protease [3], and
the structure predictions for six proteins associated with
SARS-CoV-2 using the most up-to-date version of their
AlphaFold system [4].
No matter how quickly the initial hits can be found
with or without AI, it will take long years for any such
novel molecule to progress through treacherous waters of
clinical trials and end up as medicines in doctors’ hands
at the bedside. The much needed immediate relief can
∗Correspondence and requests for materials should be addressed
to POF (email: pf@gero.ai)
only be secured by repurposing approved or clinical-stage
investigational drugs fast-tracked into clinical trials [5].
According to the report, most of the drugs currently
tested against SARS-CoV-2 are Direct-acting antiviral
agents (DAAs) and are representatives of a class of drugs
targeting viral proteins. The repurposed drugs in the
clinical trials were initially designed against HIV or in-
fluenza. A small minority of experimental treatments are
the Host-targeting antivirals (HTA) acting against host
factors, including inflammatory cytokines.
In this work, we utilized a deep neural network to
mine gene expression signatures for experimental and ap-
proved drugs with potential activity against coronavirus-
born diseases. To achieve this goal, we searched for gene
expression profiles of the molecular perturbations most
closely mimicking the gene expression signature of the
COBP2 gene knockdown. The gene is required for repli-
cation of SARS-CoV [6], which is closely genetically re-
lated to SARS-Cov-2 with 79% identity [7].
We speculated that the same gene may be necessary
for the replication of the genetically similar SARS-CoV-
2. A large number of the top-scoring molecules were al-
ready suggested for repurposing as broad-spectrum an-
tivirals based on the drug’s performance in experimental
models of viral diseases. One of the approved drugs, ni-
tazoxanide, has recently demonstrated activity against
SARS-CoV-2 [8]. According to the available literature,
a few more drug compounds from our list have been al-
ready tested and found active in models involving related
viruses in vitro, such as MERS-CoV or SARS-CoV. We,
therefore, call for characterization of the remaining top-
ranked compounds against SARS-CoV-2. All the drugs
selected in our study have been already approved for clin-
ical trials. Hence any experimental confirmation of the
activities predicted here may build the rationale for a
trial involving human subjects and, with some effort and
luck, translated into a cost- and time- effective solution
to the unfolding medical emergency.
2
RESULTS
Deep neural network for batch removal in LINCS
L1000 dataset
Deep learning with its capabilities to approximate gen-
eral non-linear dependencies of input data was recently
demonstrated as a powerful tool for gene expression data
analysis and batch removal. We trained the deep neu-
ral network (DNN) with the architecture similar to that
described in [9]. To enable identification of biologically
similar molecular and genetic perturbations, we gener-
ated compressed 20-dimensional representations (embed-
dings) of the 978-dimensional differential gene expression
signatures (LINCS L1000 data processing at Level 4 [10].
The DNN was constructed of the three blocks:
DenseNet block, Embeddings, and Perturbagen classifier,
see Fig. 1a. The DenseNet was implemented as suggested
in [9] with 32 hidden layers and a growth rate of 48. The
Embeddings layer is a dense layer with 2514 Input size
and 20 output size. The perturbagen classifier was used
to predict the perturbagen class (pert_iname label in the
L1000 dataset) from the embedding vector using addi-
tive margin Softmax (AM-Softmax) [11] with the margin
value was set to 0.2as in [9].
The training data included profiles from LINCS
PHASE I (GEO accession number GSE92742) and
PHASE II (GEO accession number GSE70138). We re-
moved samples related to cell lines in which less than
5000 profiles were measured. The preprocessed dataset
involving 1467244 gene expression profiles corresponding
to 27870 unique molecular perturbation classes was split
into the training and test datasets at 80/20 ratio. The
test dataset did not include gene expression signatures
from the training dataset.
To validate the performance of the DNN we used the
same metrics as proposed in [9]. The best model achieved
the area under the receiver operating characteristic curve
(ROC AUC) recovery of the molecular, 0.871, and ge-
netic, 0.764, perturbations, respectively. Both numbers
were significantly better than 0.656 and 0.616 for the
same molecular perturbation classes in the raw LINCS
L1000 data. The model was able to pull structurally
similar compounds (with Tanimoto similarity score bet-
ter than 0.85) with ROC AUC of 0.745.
To obtain the compressed representation of each molec-
ular perturbation class we computed consensus profiles
for each dose and time by averaging over all the available
repeats and all cell lines. For the analysis, we used the
measurements of the effects of small molecules produced
at 10µM concentration at the time point of 24h(and 96h
for the genetic perturbations).
Search for antivirals against SARS-CoV-2
A set of host factors relevant for SARS-coronavirus
(SARS-CoV) replication was identified by means of a
siRNA library screen targeting the human kinome in [6].
The depletion of the β2subunit of the coatomer protein
complex (COPB2) in the followup experiment produced
the strongest antiviral effect, reduced SARS-CoV protein
expression, and produced a >2-log reduction in virus
yield.
We assumed that the interactions between the virus
and host may be shared by the genetically similar viruses
and proceeded with the search for small molecules capa-
ble to recreate the effect of the COPB2 genetic knock-
out. Fortunately, LINCS L1000 database provides gene
expression signatures of a series of shRNA knockouts of
COPB2. We pooled the consensus signature from multi-
ple shRNA experiments and searched for molecular per-
turbations most similar to the knockout. The similarity
measure, the CoV similarity score, is the distance be-
tween the molecular perturbation and COPB2 knockout
signatures in the latent variables space (see Fig. 1b for
the distribution of the computed similarity scores).
The results of the calculations are summarized in Table
1 and divided into categories reflecting their development
status (including approved, and experimental drugs) ac-
cording to the DrugBank database. For each of the drugs,
we provided the Mechanism of Action (MoA) and the
development status (if applicable). We also performed a
literature search for evidence of antiviral activity in ex-
periments involving coronaviruses (the “direct evidence”
column) and other viral infections (“indirect evidence”
column).
Of note, the top-scored molecular perturbation was
Brefeldin A, which is a natural lactone antiviral com-
pound produced by the fungus Penicillium brefeldianum.
This is a good sanity check for the computational
pipeline since according to [12], Brefeldin A treatment
of SARS-CoV-infected cells significantly reduced replica-
tion as well as the accumulation of virus-induced mem-
brane structures (see [6] for mechanistic details relating
Brefeldin A activity and COPB2 knockout).
DISCUSSION
Direct-acting antiviral agents (DAAs) have been very
successful against viral infections in medical practice.
However, DAAs suffer from several inherent limitations,
including narrow-spectrum antiviral profiles and liability
to drug resistance. In comparison, host targeting an-
tivirals (HTAs) target host proteins that are probably
broadly required for various viral infections. This is why
HTAs are not only perceived but also demonstrated to
exhibit broad-spectrum antiviral activities. In addition,
host proteins are not under the genetic control of the viral
genome, and hence HTAs possess a much higher genetic
barrier to drug resistance as compared with DAAs (see,
e.g., [19] for the most recent review).
In this work, we focused on repurposing of approved
or clinical-stage investigational drugs against COVID-19
and other coronavirus-borne diseases. We turned to the
3
(a)
(b)
FIG. 1: a) Deep neural network (DNN) architecture employed for batch-removal and the construction of the
compressed gene expression representation; b) the distributions of the CoV score (similarity to the gene-expression
signature of COPB2 knockdown) for all, all molecular perturbations, and experimental drugs, respectively.
Broad Institute LINCS L1000 database of biological ef-
fects (gene expression signatures) of molecular and ge-
netic perturbations. The resource is commonly used for
MoA studies and drug repurposing. Unfortunately, most
of gene expression variance in transcriptomes is typically
associated with non-biological factors (commonly known
as batch effects) and hence may be of limited use without
sophisticated batch removal procedures [20].
We followed and improved on approach from [9] and
reduced batch effects with the help of a deep neural net-
work. We used the compressed low-dimensional rep-
resentations of the gene expression data from LINCS
L1000 dataset and searched for gene expression profiles
of the molecular perturbations most closely mimicking
the gene expression signature of the COBP2 gene knock-
down. The gene was identified in a full-kinome iRNA
screen and was confirmed as being required for the ge-
netically similar SARS-CoV replication in vitro [6].
Based on the computed similarity to COBP2 knockout,
we generated two lists of predicted antivirals separately
among the approved and investigational drugs. Both ta-
bles were significantly enriched by HTAs with broad ac-
tivity against viruses ranging from coronaviruses (MERS
and SARS) to influenza and HCV.
Notably, the in vitro efficacy (EC50) of the investiga-
tional drugs in models of viral diseases are most of the
time better than that of the approved drugs. This cir-
cumstance may reflect medicinal chemistry advances over
the years as we shift our attention from the “old” and al-
ready approved to the “newer” investigational drugs.
Among the approved drugs, niclosamide and nitazox-
anide are known for broad-spectrum antiviral action and
exhibited low-micromolar range activities in vitro mod-
els of coronavirus diseases. Niclosamide is approved for
the treatment of tapeworm infections and is listed in the
WHO essential medicines. Nitazoxanide, a commercial
antiprotozoal agent with an antiviral potential against
a broad range of viruses including human and animal
coronaviruses. Currently, the drug is listed in multiple
clinical trials against viral diseases. In the most recent
work, nitazoxanide inhibited the SARS-CoV-2, the virus
behind the COVID-19, at a low-micromolar concentra-
tion (EC50 = 2.12µM ;CC50 >35.53µM ; the selectivity
index (SI)>16.76) [8].
Although there is no data on the antiviral activity
of ixazomib, a proteasome inhibitor, there is evidence
that this class of compounds possesses a broad-spectrum
antiviral activity. The clinically approved proteasome
inhibitor bortezomib (PS-341) significantly reduces in-
fluenza A virus and vesicular stomatitis virus propaga-
tion at EC50 = 50 −100nM , while exhibiting some cy-
totoxicity [21]. It is hypothesized that bortezomib may
act as an antiviral agent via induction of the type I IFN
response. Authors of [22] report an EC50 of influenza
virus infection by bortezomib of 198nM in human non-
malignant RPE cells (CC50 of bortezomib for these cells
exceeds 10µM). Bortezomib also inhibits herpes simplex
virus (HSV) infection (EC50 = 4 −50nM for different
strains) since proteasomal degradation activity is criti-
cal for early steps of HSV infection [23]. Proteasome
4
TABLE I: The top-scoring drugs according to the predicted antiviral effect.
Molecule CoV-Score MoA Status Direct evidence Indirect evidence
Top LINCS
L1000
Brefeldin A Arf inhibitor N/A EC50 = 21.4µM (SARS-
CoV) [12] Herpes simplex, Newcas-
tle disease, papillomavirus
and polyomavirus [13]
Approved
drugs
Niclosamide 0.57 N/A Approved for the treat-
ment of tapeworm infec-
tions
EC50 = 3.12µM (SARS-
CoV) [14] Rhinoviruses (HRV),
influenza virus, Chikun-
gunya virus, Zika virus
[15]
Nitazoxanide 0.35 N/A Approved antiprotozoan,
currently is in clinical tri-
als for influenza (Phase
3), rotavirus or norovirus
(Phase 2), Hepatitis B and
C, HIV
EC50 = 2µM
(MERS-CoV) [16],
EC50 = 2.12µM (SARS-
CoV-2) [8]
Influenza EC50 =
0.2−1.5µM, syncy-
tial virus, parainfluenza,
rotavirus, norovirus,
hepatitis B, hepatitis
C, dengue, yellow fever,
Japanese encephalitis
virus and HIV [17]
Afatinib 0.34 EGFR Approved for the treat-
ment of non-small cell lung
carcinoma
Drugs with the same MoA
were effective in in vivo
models of SARS-Cov
N/A
Ixazomib 0.33 Proteasome
inhibitor Approved for the treat-
ment of multiple myeloma Another inhibitor with
the same MoA suppressed
replication of SARS-Cov
at EC50 <1µM
Another inhibitor with
the same MoA suppressed
replication of influenza A
virus at EC 50 <1µM
5more drugs 0.29 −0.34 Known MoA 2approved, 3withdrawn N/A Some active against Poly-
omaviruses
Reserpine 0.29 Vesicular
monoamine
transporters
antagonist
Natural compound. Used
as an antihypertensive (in
combination). Was used as
an antidepressant and to
treat dyskinesia in Hunt-
ington’s disease
EC50 = 3.4µM (SARS-
CoV) [18] N/A
Investigational
drugs
Obatoclax 0.33 Mcl-1 (Bcl-
2 family)
inhibitor,
senolytic
Phase 2 clinical trials
for leukemia, lymphoma,
myelofibrosis, and masto-
cytosis
N/A Was effective in vitro
against Influenza
A and Zika viruses
(EC50 <0.1µM )
NVP-
AUY922
0.23 Hsp90 in-
hibitor Phase 2 clinical trials, on-
cology N/A Was effective in vitro
against Influenza A and
Measles (EC50 <0.1µM )
7more drugs 0.20 −0.38 Known MoA Phase 1/2 N/A 2some active against In-
fluenza and Dengue
inhibitor MG-132 inhibits the replication of hepatitis E
and C viruses [24], cytomegalovirus [25].
Inhibition of the proteasome in sub-µM concentra-
tions by different chemical compounds, such as MG-
132 or bortezomib appeared to not only impair entry
but also RNA synthesis and subsequent protein expres-
sion of different CoVs, including mouse hepatitis virus
(MHV), feline infectious peritonitis virus, and, most no-
tably, SARS-CoV [26]. Treatment of mice with SARS-
like pneumonitis induced by murine hepatitis virus strain
1 (MHV-1) with the proteasome inhibitor PDTC (5
mg/kg daily s.c.), MG-132 (0.5mg/kg daily s.c.), or PS-
341 (or bortezomib, 0.25 mg/kg daily s.c.) led to 40%
survival (p < 0.01), with a concomitant improvement of
lung histology, reduced pulmonary viral replication, de-
creased pulmonary STAT phosphorylation, and reduced
pulmonary inflammatory cytokine expression [27].
Reserpine is another predicted “hit” and a natural com-
5
pound used as an antihypertensive (in combination with
other drugs). Previously, it was identified as a second
most active (EC50 = 3.4µM ) compound against SARS-
CoV in a screening of 10,000 compounds [18].
One of the investigational compounds in our list is the
Hsp90 inhibitor NVP-AUY922. This should not be sur-
prising, since some viruses particularly depend on the
cellular chaperoning apparatus. Accordingly, another
HSP90 inhibitor, 17-DMAG, reduced the endpoint foot
and mouth disease virus titer by more than 10-fold at
0.163µM concentration when applied prior to the infec-
tion [28]. Another HSP90 inhibitor geldanamycin ex-
hibited broad-spectrum activity against viruses includ-
ing SARS-CoV with EC50 ranging between 0.5and 4µM
(HIV-1 and SARS-CoV exhibited the highest sensitivity
to geldanamycin in [29]).
Blocking EGFR receptor kinase activity by approved
inhibitors broadly impaired infection by all major HCV
genotypes and viral escape variants in cell culture and
in a human liver chimeric mouse model in vivo [30].
In EGFR(DSK5) mice with constitutively active EGFR,
SARS-CoV infection causes enhanced lung disease [31,
32]. In a separate work involving Respiratory syncy-
tial virus (RSV) challenge model, EGFR activation sup-
presses IFN regulatory factor (IRF 1)-induced IFN-λpro-
duction and thus contributed to the viral infection. On
the contrary, EGFR inhibition during viral infection aug-
mented IRF1, IFN-λ, and decreased RSV titers [33].
Notably, many survivors of SARS-CoV and apprently
SARS-Cov-2 infections develop pulmonary fibrosis (PF),
with a higher prevalence in older patients. In mouse mod-
els of SARS-CoV pathogenesis, the wound repair path-
way, controlled by the epidermal growth factor receptor
(EGFR), is critical to recovery from SARS-CoV-induced
tissue damage.
Age, as well as pre-existing chronic diseases, are risk
factors for COVID-19 mortality. According to the pro-
portional hazards model from [34], the mortality rate
doubling time is approximately 10 years (HR = 1.10) and
is consistent with mortality rate doubling time from the
Gompertz mortality law. It is therefore intriguing to un-
derstand if there are common mechanisms through which
the aging and the prevalence of age-related diseases may
contribute to adverse outcomes.
One of the most interesting examples from our list in
relation to aging is obatoclax, which is reported to stop
influenza virus replication at a very decent 10nM con-
centration. Obatoclax is the discontinued Bcl-2 inhibitor
similar to navitoclax. Both compounds have been in-
vestigated as a senolytics [35, 36], i.e. belong to a class
of drugs reducing the number of so-called senescent cells
(SCs) [37]. The number of SCs increase with age along
with the level of inflammatory cytokines such as IL-6 as
the part of the Senescence Associated Secretory Pheno-
type (SASP). SCs are implicated in the pathogenesis of
multiple age-related diseases, including Idiopathic Pul-
monary Fibrosis. IL-6 is the target of Tosilizumab and is
now tested in clinical trials against COVID-19 [5].Since
COVID-19 mortality is reported to be associated with
cytokine storm and fibrosis (IPF), it would be interest-
ing to see if senolytic drugs (including Hsp90 inhibitors
with senolytic properties [38]) may provide benefits be-
yond reducing the virus replication rate (most probably
by stimulating apoptosis).
In conclusion, we would like to assert that the pre-
sented work does not include any experimental validation
of the predicted host-targeting antivirals. Nevertheless,
the over-representation of the known with broad antivi-
ral activities among our top-scoring compounds list is
very encouraging. We believe that the calculations pre-
sented here highlight the capabilities of modern AI/ML
pipelines for tackling the most challenging medical prob-
lems almost in real-time. The repurposing of existing
drugs from this work could lead to meaningful solutions
to the COVID-2019 pandemic and other more mundane
diseases further down the road.
We encourage academic and industrial collaborations
to validate the results of this research and further develop
the most successful compounds against COVID-19. The
regulatory and legal status of several identified drug can-
didates allow starting immediate clinical trials.
CONFLICT OF INTERESTS
KA, OB and POF are employees of Gero. PF is a
shareholder of Gero. The company develops holds I.P.
covering use of the drugs mentioned in this draft against
a broad range of viral diseases
[1] Chaolin Huang, Yeming Wang, Xingwang Li, Lili Ren,
Jianping Zhao, Yi Hu, Li Zhang, Guohui Fan, Jiuyang
Xu, Xiaoying Gu, et al. Clinical features of patients in-
fected with 2019 novel coronavirus in wuhan, china. The
Lancet, 395(10223):497–506, 2020.
[2] Yunpeng Ji, Zhongren Ma, Maikel P Peppelenbosch, and
Qiuwei Pan. Potential association between covid-19 mor-
tality and health-care resource availability. The Lancet
Global Health, 2020.
[3] Alex Zhavoronkov, Vladimir Aladinskiy, Alexander Zhe-
brak, Bogdan Zagribelnyy, Victor Terentiev, Dmitry S
Bezrukov, Daniil Polykovskiy, Rim Shayakhmetov, An-
drey Filimonov, Philipp Orekhov, et al. Potential covid-
2019 3c-like protease inhibitors designed using generative
deep learning approaches, 2020.
[4] Andrew W Senior, Richard Evans, John Jumper, James
Kirkpatrick, Laurent Sifre, Tim Green, Chongli Qin, Au-
gustin Žídek, Alexander WR Nelson, Alex Bridgland,
et al. Improved protein structure prediction using po-
6
tentials from deep learning. Nature, pages 1–5, 2020.
[5] C Harrison. Coronavirus puts drug repurposing on the
fast track. Nature Biotechnology, 2020.
[6] Adriaan H de Wilde, Kazimier F Wannee, Florine EM
Scholte, Jelle J Goeman, Peter ten Dijke, Eric J Snijder,
Marjolein Kikkert, and Martijn J van Hemert. A kinome-
wide sirna screen identifies proviral and antiviral host
factors in sars-coronavirus replication, including pkr and
early secretory pathway proteins. Journal of Virology,
pages JVI–01029, 2015.
[7] Roujian Lu, Xiang Zhao, Juan Li, Peihua Niu, Bo Yang,
Honglong Wu, Wenling Wang, Hao Song, Baoying
Huang, Na Zhu, et al. Genomic characterisation and
epidemiology of 2019 novel coronavirus: implications
for virus origins and receptor binding. The Lancet,
395(10224):565–574, 2020.
[8] Manli Wang, Ruiyuan Cao, Leike Zhang, Xinglou
Yang, Jia Liu, Mingyue Xu, Zhengli Shi, Zhihong Hu,
Wu Zhong, and Gengfu Xiao. Remdesivir and chloro-
quine effectively inhibit the recently emerged novel coron-
avirus (2019-ncov) in vitro. Cell research, 30(3):269–271,
2020.
[9] Yoni Donner, Stephane Kazmierczak, and Kristen Fort-
ney. Drug repurposing using deep embeddings of gene ex-
pression profiles. Molecular pharmaceutics, 15(10):4314–
4325, 2018.
[10] Aravind Subramanian, Rajiv Narayan, Steven M
Corsello, David D Peck, Ted E Natoli, Xiaodong Lu,
Joshua Gould, John F Davis, Andrew A Tubelli, Ja-
cob K Asiedu, et al. A next generation connectivity map:
L1000 platform and the first 1,000,000 profiles. Cell,
171(6):1437–1452, 2017.
[11] Feng Wang, Jian Cheng, Weiyang Liu, and Haijun Liu.
Additive margin softmax for face verification. IEEE Sig-
nal Processing Letters, 25(7):926–930, 2018.
[12] Kevin Knoops, Cindy Swett-Tapia, Sjoerd HE van den
Worm, Aartjan JW Te Velthuis, Abraham J Koster,
A Mieke Mommaas, Eric J Snijder, and Marjolein
Kikkert. Integrity of the early secretory pathway pro-
motes, but is not required for, severe acute respiratory
syndrome coronavirus rna synthesis and virus-induced re-
modeling of endoplasmic reticulum membranes. Journal
of virology, 84(2):833–846, 2010.
[13] Charles Grose and Daniel J Klionsky. Alternative au-
tophagy, brefeldin a and viral trafficking pathways, 2016.
[14] Chang-Jer Wu, Jia-Tsrong Jan, Chi-Min Chen, Hsing-
Pang Hsieh, Der-Ren Hwang, Hwan-Wun Liu, Chiu-Yi
Liu, Hui-Wen Huang, Su-Chin Chen, Cheng-Fong Hong,
et al. Inhibition of severe acute respiratory syndrome
coronavirus replication by niclosamide. Antimicrobial
agents and chemotherapy, 48(7):2693–2696, 2004.
[15] Wei Chen, Robert A Mook Jr, Richard T Premont, and
Jiangbo Wang. Niclosamide: beyond an antihelminthic
drug. Cellular signalling, 41:89–96, 2018.
[16] Jean-François Rossignol. Nitazoxanide, a new drug can-
didate for the treatment of middle east respiratory syn-
drome coronavirus. Journal of infection and public health,
9(3):227–230, 2016.
[17] Jean-François Rossignol. Nitazoxanide: a first-in-class
broad-spectrum antiviral agent. Antiviral research,
110:94–103, 2014.
[18] Chung-Yi Wu, Jia-Tsrong Jan, Shiou-Hwa Ma, Chih-
Jung Kuo, Hsueh-Fen Juan, Yih-Shyun E Cheng, Hsien-
Hua Hsu, Hsuan-Cheng Huang, Douglass Wu, Ashraf
Brik, et al. Small molecules targeting severe acute res-
piratory syndrome human coronavirus. Proceedings of
the National Academy of Sciences, 101(27):10012–10017,
2004.
[19] Xingyue Ji and Zhuorong Li. Medicinal chemistry strate-
gies toward host targeting antiviral agents. Medicinal
Research Reviews, 2020.
[20] Ingrid M Lönnstedt and Sven Nelander. Fc1000: nor-
malized gene expression changes of systematically per-
turbed human cells. Statistical applications in genetics
and molecular biology, 16(4):217–242, 2017.
[21] Sabine Eva Dudek, Christina Luig, Eva-Katharina Pauli,
Ulrich Schubert, and Stephan Ludwig. The clinically ap-
proved proteasome inhibitor ps-341 efficiently blocks in-
fluenza a virus and vesicular stomatitis virus propagation
by establishing an antiviral state. Journal of virology,
84(18):9439–9451, 2010.
[22] Oxana V Denisova, Laura Kakkola, Lin Feng, Jakob
Stenman, Ashwini Nagaraj, Johanna Lampe, Bhagwan
Yadav, Tero Aittokallio, Pasi Kaukinen, Tero Ahola,
et al. Obatoclax, saliphenylhalamide, and gemcitabine
inhibit influenza a virus infection. Journal of Biological
Chemistry, 287(42):35324–35332, 2012.
[23] Seth M Schneider, Suzanne M Pritchard, George A
Wudiri, Chasity E Trammell, and Anthony V Nicola.
Early steps in herpes simplex virus infection blocked by
a proteasome inhibitor. MBio, 10(3):e00732–19, 2019.
[24] Lei Xu, Xinying Zhou, Maikel P Peppelenbosch, and Qi-
uwei Pan. Inhibition of hepatitis e virus replication by
proteasome inhibitor is nonspecific. Archives of virology,
160(2):435–439, 2015.
[25] Marion Kaspari, Nina Tavalai, Thomas Stamminger, Al-
bert Zimmermann, Rita Schilf, and Elke Bogner. Pro-
teasome inhibitor mg132 blocks viral dna replication
and assembly of human cytomegalovirus. FEBS letters,
582(5):666–672, 2008.
[26] Matthijs Raaben, Clara C Posthuma, Monique H Ver-
heije, Eddie G te Lintelo, Marjolein Kikkert, Jan W
Drijfhout, Eric J Snijder, Peter JM Rottier, and Cor-
nelis AM de Haan. The ubiquitin-proteasome system
plays an important role during various stages of the coro-
navirus infection cycle. Journal of virology, 84(15):7869–
7879, 2010.
[27] Xue-Zhong Ma, Agata Bartczak, Jianhua Zhang, Ramzi
Khattar, Limin Chen, Ming Feng Liu, Aled Edwards,
Gary Levy, and Ian D McGilvray. Proteasome inhibition
in vivo promotes survival in a lethal murine model of
severe acute respiratory syndrome. Journal of virology,
84(23):12419–12428, 2010.
[28] Joseph Newman, Amin S Asfor, Stephen Berryman,
Terry Jackson, Stephen Curry, and Tobias J Tuthill. The
cellular chaperone heat shock protein 90 is required for
foot-and-mouth disease virus capsid precursor processing
and assembly of capsid pentamers. Journal of virology,
92(5):e01415–17, 2018.
[29] Yu-Huan Li, Pei-Zhen Tao, Yu-Zhen Liu, and Jian-
Dong Jiang. Geldanamycin, a ligand of heat shock pro-
tein 90, inhibits the replication of herpes simplex virus
type 1 in vitro. Antimicrobial agents and chemotherapy,
48(3):867–872, 2004.
[30] Joachim Lupberger, Mirjam B Zeisel, Fei Xiao, Chris-
tine Thumann, Isabel Fofana, Laetitia Zona, Christo-
pher Davis, Christopher J Mee, Marine Turek, Sebastian
Gorke, et al. Egfr and epha2 are host factors for hepatitis
7
c virus entry and possible targets for antiviral therapy.
Nature medicine, 17(5):589, 2011.
[31] Thiagarajan Venkataraman, Christopher M Coleman,
and Matthew B Frieman. Overactive epidermal growth
factor receptor signaling leads to increased fibrosis after
severe acute respiratory syndrome coronavirus infection.
Journal of virology, 91(12):e00182–17, 2017.
[32] Thiagarajan Venkataraman and Matthew B Frieman.
The role of epidermal growth factor receptor (egfr) signal-
ing in sars coronavirus-induced pulmonary fibrosis. An-
tiviral research, 143:142–150, 2017.
[33] April Kalinowski, Benjamin T Galen, Iris F Ueki, Ying
Sun, Arielle Mulenos, Awo Osafo-Addo, Brian Clark,
John Joerns, Wei Liu, Jay A Nadel, et al. Respiratory
syncytial virus activates epidermal growth factor recep-
tor to suppress interferon regulatory factor 1-dependent
interferon-lambda and antiviral defense in airway epithe-
lium. Mucosal immunology, 11(3):958–967, 2018.
[34] Fei Zhou, Ting Yu, Ronghui Du, Guohui Fan, Ying Liu,
Zhibo Liu, Jie Xiang, Yeming Wang, Bin Song, Xiaoying
Gu, et al. Clinical course and risk factors for mortal-
ity of adult inpatients with covid-19 in wuhan, china: a
retrospective cohort study. The Lancet, 2020.
[35] Reut Yosef, Noam Pilpel, Ronit Tokarsky-Amiel, Anat
Biran, Yossi Ovadya, Snir Cohen, Ezra Vadai, Liat
Dassa, Elisheva Shahar, Reba Condiotti, et al. Directed
elimination of senescent cells by inhibition of bcl-w and
bcl-xl. Nature communications, 7(1):1–11, 2016.
[36] Yi Zhu, Tamara Tchkonia, Heike Fuhrmann-Stroissnigg,
Haiming M Dai, Yuanyuan Y Ling, Michael B Stout,
Tamar Pirtskhalava, Nino Giorgadze, Kurt O Johnson,
Cory B Giles, et al. Identification of a novel senolytic
agent, navitoclax, targeting the bcl-2 family of anti-
apoptotic factors. Aging cell, 15(3):428–435, 2016.
[37] Bennett G Childs, Martina Gluscevic, Darren J Baker,
Remi-Martin Laberge, Dan Marquess, Jamie Dananberg,
and Jan M van Deursen. Senescent cells: an emerging tar-
get for diseases of ageing. Nature reviews Drug discovery,
16(10):718, 2017.
[38] Heike Fuhrmann-Stroissnigg, Yuan Yuan Ling, Jing
Zhao, Sara J McGowan, Yi Zhu, Robert W Brooks, Diego
Grassi, Siobhan Q Gregg, Jennifer L Stripay, Akaitz Dor-
ronsoro, et al. Identification of hsp90 inhibitors as a novel
class of senolytics. Nature communications, 8(1):1–14,
2017.
