key: cord-0753613-xw8o189s
authors: Blasiak, A.; Lim, J. J.; Seah, S. G. K.; Kee, T.; Remus, A.; Chye, D. H.; Wong, P. S.; Hooi, L.; Truong, A. T. L.; Le, N.; Chan, C. E. Z.; Desai, R.; Ding, X.; Hanson, B. J.; Chow, E. K.-H.; Ho, D.
title: IDentif.AI: Artificial Intelligence Pinpoints Remdesivir in Combination with Ritonavir and Lopinavir as an Optimal Regimen Against Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2)
date: 2020-05-08
journal: nan
DOI: 10.1101/2020.05.04.20088104
sha: d1a5f5601ec8e1e2b04ffb93d461ae6c842e6ded
doc_id: 753613
cord_uid: xw8o189s

The emergence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and coronavirus disease 2019 (COVID-19) has led to the rapid initiation of urgently needed clinical trials of repurposed drug combinations and monotherapies. These regimens were primarily relying on mechanism-of-action based selection of drugs, many of which have yielded positive in vitro but largely negative clinical outcomes. To overcome this challenge, we report the use of IDentif.AI, a platform that rapidly optimizes infectious disease (ID) combination therapy design using artificial intelligence (AI). In this study, IDentif.AI was implemented on a 12-drug candidate therapy search set representing over 530,000 possible drug combinations. IDentif.AI demonstrated that the optimal combination therapy against SARS-CoV-2 was comprised of remdesivir, ritonavir, and lopinavir, which mediated a 6.5-fold improvement in efficacy over remdesivir alone. Additionally, IDentif.AI showed hydroxychloroquine and azithromycin to be relatively ineffective. The identification of a clinically actionable optimal drug combination was completed within two weeks, with a 3-order of magnitude reduction in the number of tests typically needed. IDentif.AI analysis was also able to independently confirm clinical trial outcomes to date without requiring any data from these trials. The robustness of the IDentif.AI platform suggests that it may be applicable towards rapid development of optimal drug regimens to address current and future outbreaks.

IDentif.AI showed hydroxychloroquine and azithromycin to be relatively ineffective. The identification of a clinically actionable optimal drug combination was completed within two weeks, with a 3-order of magnitude reduction in the number of tests typically needed. IDentif.AI analysis was also able to independently confirm clinical trial outcomes to date without requiring any data from these trials. The robustness of the IDentif.AI platform suggests that it may be applicable towards rapid development of optimal drug regimens to address current and future outbreaks.

Drug repurposing, or the use of approved and investigational therapies for other indications, has been a widely implemented strategy towards treating COVID- 19 . Examples include clinical studies of ritonavir and lopinavir (1); hydroxychloroquine in combination with azithromycin (2) ; favipiravir in combination with tocilizumab (NCT04310228); remdesivir (3); and losartan (NCT04312009), among others. In one trial, remdesivir met trial endpoints, reducing the median time to recovery from 15 days to 11 days (P < 0.001), and has ultimately received United States Food and Drug Administration (FDA) authorization for emergency use in severe COVID-19 patients (4) . The majority of trial outcomes are either pending or have not shown clinical benefit over standard of care or placebo. As such, while drug repurposing enables rapid intervention against COVID-19, there is still a lack of clarity with regards to how to best treat this disease.

Traditional methods for implementing combination therapy and monotherapy based on drug repurposing rely on mechanism of action (MOA)-based drug selection and standard clinical dosing guidelines to achieve drug synergy and therapeutic efficacy. For example, a recent preclinical study showed that remdesivir as well as high-dose chloroquine were efficacious towards SARS-CoV-2 in vitro. While this is an established approach that has led to promising candidate therapies, many of these regimens were not able to translate their in vitro outcomes into successful clinical results. Therefore, optimal efficacy that is clinically relevant is a different objective that presents substantial challenges to traditional drug screening and repurposing All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint methods. For example, if candidate effective drugs are given in combination at suboptimal respective doses, resulting efficacy is moderate or even absent. At the same time, the relative doses between drugs within a combination can substantially impact treatment efficacy and toxicity due to unpredictable drug interactions. Therefore, drug dosing has a critical role in identifying which drugs belong in the optimal combination in the first place. Therefore, optimizing treatment outcomes, particularly in combination therapy, ultimately relies on selecting the right drugs at the right respective doses (5, 6) . Reconciling drug-dose parameters also requires leveraging unpredictable drug interactions in order to mediate maximal efficacy of combination therapies.

Unfortunately, simultaneously pinpointing these parameters is an extraordinarily complicated task. For example, a parameter space of 1 trillion (10 12 ) possible combinations would be created from a pool of only 12 candidate therapies interrogated at 10 dose levels. This is an insurmountable barrier for traditional drug screening. Important studies have previously sought to leverage drug synergy interactions to predict multi-drug combinations (7) . Other strategies have investigated higher order drug interactions to develop antimicrobial drug combinations (8) .

Bridging these findings with clinical validation remains a challenge due to the size of the experimental search space.

In this study, we sought to overcome these challenges in developing effective combination therapies against SARS-CoV-2 infection using the IDentif.AI platform. IDentif.AI harnesses a quadratic relationship between therapeutic inputs (e.g. drug and dose) and biological outputs (e.g. quantifiable measurements of efficacy, safety) to experimentally pinpoint optimal combinations from large parameter spaces with a marked reduction in the number of required experiments (Fig.   1 ). Identif.AI does not use pre-existing training datasets, but rather uses an orthogonally-designed set of calibrating regimens to simultaneously identify effective drugs and corresponding doses that optimize treatment outcomes from prohibitively large drug-dose parameter spaces that cannot be reconciled by brute force drug screening (5, 9) . In effect, IDentif.AI leverages these calibrating regimens to crowdsource SARS-CoV-2 live virus responses to experimentally drive All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint the efficacy towards an optimal outcome. In this study, IDentif.AI was applied to a 12-drug set of candidate therapies to pinpoint clinically actionable combination therapy regimens against the live SARS-Cov-2 virus isolated from a nasopharyngeal swab of a patient in Singapore (10) . The 12drug set included a broad spectrum of repurposed agents that are currently being evaluated in clinical studies for treatment of COVID-19 or being administered in conjunction with these therapies, including remdesivir (RDV), favipiravir (FPV), ritonavir (RTV), lopinavir (LPV), oseltamivir (OSV-P), dexamethasone (DEX), ribavirin (RBV), teicoplanin (TEC), losartan (LST), azithromycin (AZT), chloroquine (CQ), and hydroxychloroquine (HCQ). Based on prior studies of minimal resolution experimental design, 3 dosing levels were employed with these 12 drugs, creating a combinatorial space of 531,000 regimens (11) . With a 3-order of magnitude reduction in required tests, we identified a clinically actionable list of 2-,3-, and 4-drug combinations ranked based on viral inhibition efficacy with accompanying safety data against kidney epithelial cells (Vero E6), liver epithelial cells (THLE-2) and cardiomyocytes (AC16). The top-ranked combination was comprised of remdesivir, ritonavir, and lopinavir, which mediated a 6.5-fold increase in efficacy (viral inhibition %) compared to remdesivir alone. Further demonstrating the clinical actionability of IDentif.AI, hydroxychloroquine and azithromycin combination was shown to be a relatively ineffective regimen, mirroring recent clinical results. Importantly, the IDentif.AIpinpointed relative efficacy of the combinations and monotherapies was independently confirmatory of many of the clinical trial endpoints to date. These outcomes, coupled with the fact that foundational precursors to IDentif.AI have been clinically validated for infectious disease, oncology, and organ transplantation human studies, support the potential application of IDentif.AI as a clinical decision support platform for the optimized design of actionable combination therapy regimens (12) (13) (14) .

All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

IDentif.AI, a dynamic optimization AI-based platform, identifies the drug-dose parameter space by harnessing the quadratic relationship between biological responses to external perturbations, such as drug/dose inputs (15) . IDentif.AI analysis of the drug-dose parameter space identifies drug-drug interactions and ranks optimal drug-dosage combinations. This study aimed to use IDentif.AI to determine effective optimal drug-dosage combinations from a diverse All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. IDentif.AI analysis correlated drug combinations experimental results into a second-order quadratic series. Each independent drug combination inhibition and monotherapy inhibition replicate was used in the optimization process. The second-order quadratic model is as follows:

where y represents the desired biological response output (%Inhibition), xn is the n-th drug concentration, β0 is the intercept term, β n is the single-drug coefficient of the n-th drug, β mn is the interaction coefficient between the m-th and n-th drugs and β nn is the second-order coefficient for the n-th drug, while m ≠ n. This second-order quadratic analysis and parabolic response surface plot analysis were conducted using the built-in "stepwiselm" function in Matlab R2020a

(Mathworks, Inc.) with custom-written code. IDentif.AI derived four quadratic series using bidirectional elimination approach with the P value from the F-Statistic as the selection criterion for the experimental results: %Inhibition, %Cytotoxicity, %Cytotoxicity AC16, and %Cytotoxicity All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint THLE-2. Residual-based outlier analysis was performed for all four IDentif.AI series. Single replicates identified as outliers remained in the data set to account for biological variation. The combinations with all replicates identified as outliers were excluded from the data set and the IDentif.AI analysis was repeated.

IDentif.AI analysis yielded both drug-drug interaction plots and optimized drug combinations. The optimized drug combinations were ranked according to corresponding %Inhibition from the correlated second-order quadratic series with the %Cytotoxicity of the celllines (Vero E6, AC16, and THLE-2) serving as qualitative indicators for consideration. The predictive power was also calculated via adjusted R 2 to establish the robustness of IDentif.AI optimization considering the number of drug and drug-drug interaction terms. Correlation coefficients were derived from the experimental output values and projected output values for the corresponding drug combinations.

All experiments were performed in at least triplicate biological repeats with data presented as means ± standard deviation (SD), unless otherwise stated. Shapiro-Wilk normality test was used to determine if samples were from normally distributed populations. Variance equality was tested with Bartlett's test. The Kruskal-Wallis test by ranks was used for multiple comparisons, followed by Dunn's post hoc test for pairwise comparisons. Student's two-tailed t test and Wilcoxon rank sum test were used for comparing individual samples from normally and nonnormally distributed populations, respectively. Bonferroni post-hoc correction was applied to account for multiple comparisons. Statistical analyses for coefficient estimation in the IDentif.AI analyses were performed using sum of squares F-test. Alongside the P-values, the results were interpreted in the light of logic, background knowledge and the specifics of the experimental design (16) . All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Only high concentrations (>1 µM) of RDV, LPV, CQ and HCQ achieved half maximal absolute effective concentration (EC50) for the viral inhibition within the tested concentration ranges. High concentrations (>20 µM) of RTV, LPV and CQ led to half maximal absolute cytotoxic All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint concentration (CC50) within the tested concentration ranges (Table 1) . These results indicated low cellular effects of the selected monotherapies at the tested concentrations. No effect of the maximum vehicle concentration (0.1% DMSO) was detected on viral inhibition or on cytotoxicity (Student's t-test, n = 12, P > 0.05). The EC50 and CC50 of HCQ, CQ, RDV, FPV, and RBV were different from previously reported values, attributable to differences in the experimental conditions (e.g. SARS-CoV-2 strain, assays, incubation periods) (24, 25) . Because a common source of failure in translating in vitro results to clinical trials, including for coronaviruses, is the high ratio of EC50 to maximum plasma concentration (Cmax) achieved in the human body, Cmax was included as a crucial consideration for selecting drug concentrations at Level 1 and Level 2 for each drug that ensure none of the drugs were overrepresented in relation to other drugs and to human pharmacokinetics (Table 1, Supplemental Results) (26) . Regardless of the monotherapy antiviral activity, all drugs were considered for the combinatorial optimization process in order to identify possible unpredictable drug interactions that could markedly impact treatment efficacy and safety. In order for IDentif.AI to determine optimized drug combinations from this 12-drug set, 100

drug-dose combinations were generated according to OACD (Table S1 ) and, together with drug monotherapies at concentration Level 1 and Level 2, were evaluated for their antiviral and All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

Utilizing the single drug and OACD drug treatment data, IDentif.AI analysis determined RDV/RTV/LPV to be the most efficacious 3-drug combination. It was also present in all top 10 ranked 4-drug combinations. RDV/LPV was the top ranked 2-drug combination ( Table 2) . While RDV was identified as the most efficacious single drug, in line with current clinical trial outcomes, IDentif.AI analysis determined that the 3-drug combination of RDV/RTV/LPV is critical for achieving maximal therapeutic efficacy. IDentif.AI analysis allows for comparative ranking of all possible combinations within the 12-drug set, including analysis of regimens currently being clinically investigated but that are not observed as top ranked optimized drug combinations. Both LPV/RTV (Kaletra) and HCQ/AZT have been clinically evaluated as potential treatments against SARS-CoV-2 infection with discouraging outcomes. IDentif.AI analysis of these combinations revealed that they were identified to be sub-optimal -LPV/RTV ranked 1261 and HCQ/AZT ranked 5161 amongst all 9968 drug combinations that include up to 4-drugs, and predicted viral inhibition efficacies of 23% and 2%, respectively. The aforementioned findings were based on the IDentif.AI quadratic series assessing the %Inhibition experimental data with a close proximity as indicated by adjusted R 2 of 0.898 (Table S2) .

Multi-parameter IDentif.AI analysis allowed cytotoxicity of ranked combinations to be interrogated as well via deriving %Cytotoxicity quadratic series (Table S3- was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint and higher %Cytotoxicity in the THLE-2 cells. This IDentif.AI-derived THLE-2 %Cytotoxicity was predicted to decrease with the addition of DEX in the top 4-drug combination ( Table 2) . Outlier analysis performed for each IDentif.AI quadratic series (Fig. S2-S5 ) identified and excluded OACD combination 15 from the AC16 %Cytotoxicity data set (Fig. S4 ) and combination 46 from the THLE-2 %Cytotoxicity data set (Fig. S5) . These data sets were subsequently re-analyzed ( Fig. S6-S7) . Taken together, IDentif.AI identified RDV-based treatments as likely the most effective therapies against SARS-CoV-2 infections, with RDV/RTV/LPV capable of achieving maximal efficacy with potential reductions in overall toxicity if complemented with the fourth drug.

All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

Validation results were interpreted considering not only the P-values, but also the logic, background knowledge and specifics of the experimental design (16) . The studies confirmed

IDentif.AI ranking of RDV in combination with LPV and RTV as the optimal combination of the study, resulting in complete viral inhibition ( Fig. 2A, Table S6 ). This combination resulted in a 6.5fold increase in efficacy compared to RDV alone. RDV was confirmed as an essential driver of was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. (NCT04362332) did not induce as much viral inhibition as compared to RDV alone. These data confirm that IDentif.AI can accurately reflect the unsatisfactory outcomes observed in those clinical trials, without incorporating any prior clinical data or drug mechanism assumptions as input.

All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint Drug-drug interaction analysis of IDentif.AI results were also compared to experimental observations and known clinical investigations. %Inhibition IDentif.AI response surface plot mirrored well-documented and experimentally confirmed synergy between RTV and LPV (Fig.   3A ). In contrast, IDentif.AI identified an antagonistic interaction between RTV and OSV-P (Fig.   3B ), a combination that is currently being investigated in clinical trials (NCT04303299).

Combining RDV with LPV only, which to our knowledge has not been explored clinically as a registered trial, doubled their individual viral inhibition when added together. Accordingly, the corresponding %Inhibition IDentif.AI response surface plot identified a previously unknown synergistic interaction between RDV and LPV (Fig. 3C) . Further confirming IDentif.AI rankings and validation experiments, the RDV/RTV interaction was not significant, but when given in 3drug combination, RTV boosted the RDV/LPV interaction almost two times (Fig. 3D) . These results further highlight the ability of IDentif.AI to leverage unexpected drug-dose interactions to identify optimal drug combinations from a massive drug-dose search space.

All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . 

This study harnessed the IDentif.AI platform to interrogate a 12 drug-dose parameter space against the SARS-CoV-2 live virus to develop actionable and optimized combination therapy regimens. IDentif.AI implementation addresses several important factors when designing multi-drug regimens that are best suited for clinical translation from in vitro validation, especially under urgent scenarios like COVID-19. Importantly, IDentif.AI considers the critical need for simultaneous reconciliation of drug composition and dosing within combination therapy design.

MOA-based drug selection alone followed by dose finding, while an established method of combination therapy design, presents substantial barriers to the optimization process since drug dosing also plays a role in determining which drugs belong in an ideal combination. In lieu of validating a small number of MOA-based potential drug combinations for efficacy which is All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint commonly observed in traditional workflows, IDentif.AI takes an MOA-agnostic approach to efficiently analyze crowdsourced therapeutic responses to the live virus following expansive drugdose exposure to both outline the drug-dose space and define resulting drug-dose compositions of the optimal regimens (41) . With this data, IDentif.AI is able to leverage on unexpected dosedependent drug interactions to mediate improved treatment outcomes over MOA-based drug selection followed by dose finding.

Another critical aspect of IDentif.AI is that the in vitro drug dosing parameter space interrogated in this study is a departure from traditional drug screening approaches. In traditional drug screening, compounds that do not elicit at least a low micromolar EC50 treatment response during drug dose-response evaluations are typically removed from further consideration, thereby markedly reducing the number of candidate therapies and possible drug combinations. The removal of these drug candidates is a key driver of sub-optimal treatment responses as it ignores a broad spectrum of potential combinations that can be assessed. Lack of monotherapy efficacy does not preclude the use of these drug candidates from IDentif.AI's combinatorial search space.

Instead, IDentif.AI's approach allows for continued evaluation of these drugs to determine if they are vital toward driving previously unknown drug interactions that optimize combinatorial treatment outcomes. In addition to being observed in this study, this phenomenon has also been observed with our prior clinical studies in chronic infectious diseases and blood and solid cancers, among other indications (5, 8, 9, 42) .

The outcome of applying IDentif.AI towards combating SARS-CoV-2 infection is an extensive list of combinations ranked by efficacy and/or safety that can be queried by a clinician based on clinically actionable criteria. These include, but are not limited to: highest ranked 2-, 3-, 4-drug combinations by efficacy; highest ranked combinations that do not contain certain drugs due to supply shortages; highest ranked combinations that do not contain certain drugs or contain lower dosages of certain drugs due to patient co-morbidities; and highest ranked combination comprised of only approved therapies, among others. In the context of optimized regimen design, All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . which assesses regimen performance from the entire landscape of possible drug/dose parameters, IDentif.AI-enabled comparative evaluation of the relative efficacy of a broad spectrum of optimized regimens and clinically investigated regimens also independently confirmed the reported outcomes of clinical trials. This provides additional support for the potential application of IDentif.AI as a clinical decision support platform. For example, the relatively low efficacy exhibited HCQ alone (3.9%) or by HCQ and AZT (3% inhibition) in this study aligned with recent reporting of clinical outcomes for this drug given in mono-and combinatory therapy (39, 40). The relatively low efficacy (3.9% and 5.2%) of RTV and LPV combination when assessed by IDentif.AI at two different dosing ratios also aligned with recently reported outcomes showing no benefit over standard care (1). IDentif.AI also revealed a relatively low efficacy of FPV monotherapy (1.9% inhibition) and various combinations. This was consistent with clinical findings of FPV being potentially clinically effective only when administered with interferon-alpha, not included within our drug library (43) . Of note, RDV alone resulted in the highest relative efficacy for monotherapy (15.5%) in this study. To date, compassionate use of RDV resulted in clinical improvement of 68% of the patients, and a larger study resulted in a statistically significant improvement in median time to recovery from 15 days to 11 days (4). At the same time, an RDV study in severe COVID-19 patients was also recently terminated early (NCT04257656). Nonetheless, it has received FDA authorization for emergency use in severe COVID-19 patients. The substantial difference in efficacy observed between sub-optimal and optimal regimens highlights the importance of leveraging platforms such as IDentif.AI to systematically design combination therapies. This capability, along with the potentially predictive capacity of IDentif.AI for clinical trial outcomes could provide clinicians with an expanded arsenal of evidence-based candidate treatments and important insights into which potential treatments to further evaluate or potentially avoid under time-sensitive circumstances.

It is important to note that the results reported here are derived from primarily an in vitro SARS-CoV-2 study. Further clinical validation of the outlined combinations in randomized All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint controlled trials will be needed. It should also be noted that, while RDV did not mediate a significant clinical benefit in severe COVID-19 patients, its efficacy in patients with varying disease burden severities should be evaluated further. Furthermore, the mixed reported clinical outcomes support the need for improved regimen design of RDV-based treatment. In the event of downstream clinical validation of IDentif.AI-designed combinations, the drug dosage ratios within the combination may vary from those pinpointed by IDentif.AI. In addition, it is possible that the optimal drug combinations may vary between patients due to their severity of infection, comorbidities, and other factors. It is for these reasons that potential downstream trials may be effective at determining the potential clinical benefit of the IDentif.AI-designed combinations if the enrolled patients are stratified by these aforementioned clinical parameters. The 12-drug search set used in this study did not include every therapeutic option currently under clinical investigation.

Additional studies, including other repurposed compounds, may yield additional highly ranked and effective combination regimens. Also, as IDentif.AI can be applied to novel small molecules and antibody therapies, their inclusion into the drug pool would add further insight into other potentially actionable regimens. Furthermore, given the rapid mutagenicity of RNA viruses like SARS-CoV-2, future studies with different drug candidates and different SARS-CoV-2 strains may yield different combinations. However, the efficiency and deterministic nature of IDentif.AI allows it to derive a ranked list of optimal regimens from a given set of drug candidates against a defined in vitro infectious disease model within two weeks. This further supports its potential application as a clinical decision support platform for the optimized design of combination therapy regimens against multiple SARS-CoV-2 strains as well as future unknown pathogens that will again require rapid mobilization and clinical guidance for effective treatment options.

A.B., T.K., L.H. X.D., E.K.C., and D.H. are co-inventors or previously filed pending patents on artificial intelligence-based therapy development.

All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.04.20088104 doi: medRxiv preprint

A Trial of Lopinavir-Ritonavir in Adults Hospitalized with Severe Covid-19

Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an open-label non-randomized clinical trial

Compassionate Use of Remdesivir for Patients with Severe Covid-19

NIH clinical trial shows Remdesivir accelerates recovery from advanced COVID-19, (NIH), access date

Project IDentif.AI: Harnessing Artificial Intelligence to Rapidly Optimize Combination Therapy Development for Infectious Disease Intervention

Artificial intelligence in cancer therapy

Prediction of multidimensional drug dose responses based on measurements of drug pairs

Enhanced identification of synergistic and antagonistic emergent interactions among three or more drugs

Optimizing drug combinations against multiple myeloma using a quadratic phenotypic optimization platform (QPOP)

Epidemiologic Features and Clinical Course of Patients Infected With SARS-CoV-2 in Singapore

Maximizing Efficiency of Artificial Intelligence-Driven Drug Combination Optimization through Minimal Resolution Experimental Design

Modulating BET Bromodomain Inhibitor ZEN-3694 and Enzalutamide Combination Dosing in a Metastatic Prostate Cancer Patient Using CURATE.AI, an

Individualizing liver transplant immunosuppression using a phenotypic personalized medicine platform

Application of an ex-vivo drug sensitivity platform towards achieving complete remission in a refractory T-cell lymphoma

Systematic quantitative characterization of cellular responses induced by multiple signals

It's time to talk about ditching statistical significance

Coronavirus puts drug repurposing on the fast track

Clinical Pharmacology Perspectives on the Antiviral Activity of Azithromycin and Use in COVID-19

Pharmacologic Treatments for Coronavirus Disease

Glycopeptide Antibiotics Potently Inhibit Cathepsin L in the Late Endosome/Lysosome and Block the Entry of Ebola Virus, Middle East Respiratory Syndrome Coronavirus (MERS-CoV), and Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV)

On the use of corticosteroids for 2019-nCoV pneumonia

Considerations for Certain Concomitant Medications in Patients with COVID-19

Remdesivir and chloroquine effectively inhibit the recently emerged novel coronavirus (2019-nCoV) in vitro

In Vitro Antiviral Activity and Projection of Optimized Dosing Design of Hydroxychloroquine for the Treatment of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2)

Coronaviruses -drug discovery and therapeutic options

Remdesivir: Summary on Compassionate Use

Norvir: EPAR -Scientific Discussion

Pharmacokinetics of adjusted-dose lopinavir-ritonavir combined with rifampin in healthy volunteers

Impact of ribavirin plasma level on sustained virological response in patients treated with pegylated interferon and ribavirin for chronic hepatitis C

Safety, efficacy and pharmacokinetic evaluations of a new coated chloroquine tablet in a single-arm open-label non-comparative trial in Brazil: a step towards a user-friendly malaria vivax treatment

Food and Drug Administration, (FDA)

Food and Drug Administration, (FDA)

Pharmacokinetics of orally administered oseltamivir in healthy obese and nonobese Thai subjects

Pharmacokinetics and biochemical efficacy after single and multiple oral administration of losartan, an orally active nonpeptide angiotensin II receptor antagonist, in humans

Therapeutic drug concentrations of teicoplanin in clinical settings

Hemady: Highlights of Prescribing Information

No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint

A pilot study of hydroxychloroquine in treatment of patients with common coronavirus disease-19 (COVID-19)

No evidence of rapid antiviral clearance or clinical benefit with the combination of hydroxychloroquine and azithromycin in patients with severe COVID-19 infection

Addressing COVID-19 Drug Development with Artificial Intelligence

Output-driven feedback system control platform optimizes combinatorial therapy of tuberculosis using a macrophage cell culture model

Experimental Treatment with Favipiravir for COVID-19: An Open-Label Control Study. Engineering (Beijing)

Figs. S1-S7Tables S1-S6References 1-23