key: cord-0818073-zh5a16ky authors: Hiremath, Chaitanya N. title: Abbreviated Profile of Drugs (APOD): modeling drug safety profiles to prioritize investigational COVID-19 treatments date: 2021-07-28 journal: Heliyon DOI: 10.1016/j.heliyon.2021.e07666 sha: 9729eb410dadf87f7b2cc23b9293407712eb529a doc_id: 818073 cord_uid: zh5a16ky Safe and effective oral formulation of a drug, that is easy to store, transport, and administer, is imperative to reach the masses including those without adequate facilities and resources, in order to combat globally transmitted coronavirus disease 2019 (COVID-19). In this decision analytic modeling study, the safety of investigational COVID-19 drugs in clinical trials was assessed using the Abbreviated Profile of Drugs (APOD) methodology. The method was extensively tested for various unbiased datasets based on different criteria such as drugs recalled worldwide for failing to meet safety standards, organ-specific toxicities, cytochrome P450 inhibitors, and Food and Drug Administration (FDA) approved drugs with remarkable successes. Experimental validation of the predictions made by APOD were demonstrated by comparison with a progression of multiparametric optimization of a series of cancer drugs that led to a potent drug (GDC-0941) which went into the clinical development. The drugs were classified into three categories of safety profiles: strong, moderate and weak. A total of 3556 drugs available in public databases were examined. According to the results, drugs with strong safety profiles included molnupiravir (EIDD-2801), moderate safety profiles included dexamethasone, and weak safety profiles included lopinavir. In this analysis, the physicochemical-pharmacokinetic APOD fingerprint was associated with the drug safety profile of withdrawn, approved, as well as drugs in clinical trials and the APOD method facilitated decision-making and prioritization of the investigational treatments. An oral formulation, like a tablet or a capsule, is preferred to an injectable because it is durable, cost-effective, and easy to administer when considering the global pandemic such as An ideal oral drug is one that is rapidly and completely absorbed from the alimentary canal, explicitly distributed to its site of action in the body, metabolized in a way that does not instantly remove its activity, and eliminated suitably, without causing any harm (Hodgon, 2001) . It is estimated that nearly half of the drugs in the development fail to make it to the market due to poor safety (Hodgon, 2001) . The laboratory techniques used to determine drug safety are timeconsuming. The tell-tale signs of drug safety are in the actual makeup of the drug itself. Finding these characteristics requires a careful and exhaustive search of strong experimental evidence reported in scientific literature. Applying this critical information is yet another challenge. An ideal drug also has a strong safety profile and high biological activity (Singal et al., 2014) . The safety profile is usually determined using pharmacokinetics by studying the movement of the drug in the body including in the processes of absorption, distribution, metabolism, and excretion of drugs. Drugs with a strong safety profile have fewer side effects. An example is penicillin, which is virtually nontoxic, even in large doses (Wilkowske, 1977) . Drugs with a weak safety profile have adverse side effects. An example is mibefradil (posicor), which was a drug used for the treatment of hypertension and was withdrawn from the market within a year after it was first released following a spate of dangerous/lethal effects (CDER, 2004) . Drugs can produce physicological effects by initiating, inhibiting, modulating, or enhancing the basal biological activity in cells. Drug activity relates to its pharmacodynamics, the ability to bind to its target, and its dose. In this sense, a highly effective drug can cause a change in the behavior of the cell in a given tissue with only a small amount of the drug. On the other hand, an ineffective drug does not change the behavior of the cell or a large amount of the drug may be required to see a comparable change. A risky drug with adverse side effects has a weak safety profile, no matter how effective the drug is. An undesirable drug has a weak safety profile and low biological activity. Naturally, undesirable drugs should be identified early and eliminated promptly in the drug discovery process. Laboratory techniques routinely used in the evaluation of the safety profiles of drugs are the major cause of slowdown in the evaluation process. These techniques include enzyme assays, cell culture, translocation, immunoblotting, and phosphoprotein immunoassay on cell lines, among others (Raynaud et al., 2009 ). They should be limited to as small a number as possible of promising drugs. The APOD method allows a rapid evaluation of drug safety in a unified and cohesive way. It has been described in detail in an earlier article (Hiremath, 2007 ). Lipinski's rule of five (Ro5) with four conditions (Lipinski et al., 1997) was not adequate in revealing the drug-like characteristics of compounds. In the spirit of number five, Hiremath's rule (Hiremath, 2007) included the fifth condition with comprehensive polar surface area less than 150 Å 2 to Ro5. Briefly, APOD uses the drug's chemical properties for a mathematical model to predict its J o u r n a l P r e -p r o o f pharmacokinetic properties. In this approach, all the propertiesphysicochemical, biological, pharmacokineticwere on the same scale from 0 to 9, thereby making any direct comparison possible. Analogous to the identification of high blood pressure as a significant risk factor for heart disease, this method relies on the vast number of proven studies and a comprehensive knowledge base (Lipinski et al., 1997; Lombardo et al., 2003; Pajouhesh and Lenz, 2005; Stenberg et al., 2001; Waterbeemd et al., 2001) . For example, it is known that higher lipophilicity of compounds leads to increased metabolism and poor absorption, along with an increased probability of binding to undesired hydrophobic macromolecules, thereby increasing the potential for toxicity (Pajouhesh and Lenz, 2005) . Such interplay of chemical and pharmacokinetic properties has been integrated into the APOD approach. Currently, the ADMET-related in silico models predict tens to thousands of descriptors in different representations and units leaving a user in a maze of properties with ambiguous and indecisive outcomes (Gola et al., 2006) . Machine learning-based methods, such as PrOCTOR and TargeTox, that are target-driven drug toxicity prediction methods rely on biological interaction networks which can limit their effectiveness due to incomplete information about all possible bound proteins (Gayvert et al., 2016; Lysenko et el., 2018) . The challenges of applying machine learning lie primarily with the lack of interpretability and repeatability of its results, which may limit their application (Vamathevan et al., 2019) . With machine learning algorithms prediction, there is always a concern with overfitting or underfitting (Patel et al., 2020) . On the other hand, the wider knowledge-based APOD methodology uses the most critical properties and provides a clear verdict (Hiremath, 2007) . Holistically, APOD provides biological insight not only into absorption, distribution, metabolism, excretion, and toxicity of a drug but also its exact placement in the spectrum of drug safety profiles. J o u r n a l P r e -p r o o f SwissADME is a web tool to predict physicochemical properties, pharmacokinetics and drug-likeness (Diana, Michielin, & Zoete, 2017). It provides easy input and interpretation but it does not predict toxicity. Additionally, APOD is easy to use and allows for prediction all ADMET properties and prioritization of molecules. There are numerous commercial computational tools available to predict ADMET behavior (Wu et al., 2020) . Quantitative structure-activity relationship (QSAR) employs mathematical models to describe relationships between molecular structures and their biological activities. Although the use of QSAR models has made considerable progress in ADMET prediction, it is limited by its model expansion capability, and large experimental data are always needed for model construction; the narrow data distribution may induce over fitting and lead to inaccurate prediction results (Wu et al., 2020) . On the contrary, APOD only uses readily available physicochemical data of the drug and it also offers model expansion capability. This decision analytical model study explores the association of physicochemicalpharmacokinetic fingerprint with the drug safety profile for the COVID-19 related drugs in clinical trials and assesses whether Abbreviated Profile of Drugs (APOD) method can facilitate decision-making and prioritization of the investigational treatments. The molecular mechanisms of action and pharmacodynamics are beyond the scope of this study. The Hiremath "rule of five" are a set of five rules that the properties of a compound should satisfy to be drug-like (Hiremath, 2007) . First, the molecular weight (W) should be less than 500 daltons. Second, the number of hydrogen-bond acceptors/receivers (R) should be less where, W is the molecular weight, R is hydrogen-bond acceptors (receivers), G is the hydrogenbond donors (givers), L is the octanol-water partition coefficient, and S is the comprehensive polar surface area. The lower limit for most properties is zero because the values for properties such as molecular weight, hydrogen-bond acceptors and donors, and comprehensive polar surface area cannot be negative. However, for some compounds the partition coefficient can be negative. Based on the physicochemical knowledge base, each property is expressed in A-POD format by dividing the actual value of the compound property by its upper limit and then J o u r n a l P r e -p r o o f multiplying by the number ten. In order to obtain a single number, the normalized value must then be rounded off ("abbreviated") to the lowest integer. The pharmacokinetic knowledge base as a function of the physicochemical properties is best represented by where, A is the absorption, D is the distribution, M is the metabolism, E is the excretion, and T is the toxicity. The number represents the weight assigned to the property, and the positive and negative signs indicate direct and inverse correlations, respectively. Empirically, the APOD pharmacokinetic estimation based on the pharmacokinetic knowledge base Y is performed by the equation where, П is the APOD pharmacokinetic property, Г is the APOD chemical property, i is the number of chemical properties, α is the favorability factor (2 if favorable, 1 if unfavorable), β is the weighting factor, and Σ is summation. The APOD indicator knowledge base as a function of the pharmacokinetic properties is currently represented by J o u r n a l P r e -p r o o f Empirically, the APOD score is computed using the equation where, Ω is the APOD indicator, П is the APOD pharmacokinetic property, k is the number of pharmacokinetic properties, α is the favorability factor (2 if favorable, 1 if unfavorable), β is the weighting factor, and Σ is summation. It is important to note that the knowledge base can be easily modified at any time to update, improve, and evolve. The pharmacokinetic knowledge base can be customized to specific needs, such as intravenous route, organ-specific toxicities, etc. The immediate goal for any pharmaceutical formation is to get the absorbed drug into the bloodstream. For an oral route, the drug should be able to first survive the low pH of the gastrointestinal tract, and then diffuse across the cell membrane (lipid bilayer) which limits the molecular size. When a drug is intravenously administered, the customization of the knowledge base could involve allowing for lower lipophilicity and higher molecular size with appropriate weights based on their relative importance, for example. The power of the rule-based APOD method is due to the fact that it An interactive web tool with easy input and interpretation of results is freely available through the website https://apodvision.com/apod/apod.pl. In the APOD approach (Hiremath, 2007) , the physicochemical properties are normalized with respect to the upper limit of the corresponding properties. The resulting value for each property is multiplied by the number ten and then rounded off ("abbreviated") to the lowest integer. The property values that are violated are set to the upper limit 9 in the APOD representation. The biological activity (B) is the concentration of the drug needed to produce a biological effect on a specific target. These values are different in various targets such as humans, rats, mice, rabbits, and monkeys. The input for biological activity is in nanomolar units, accepting values from 0 to 9999. So, the highest value in biological activity correspond to lowest efficacy and the lower values in biological activity correspond to higher efficacy of drug. In contrast, the increase in the APOD value represents the increase in biological activity or efficacy of the drug. The inverse relationship is converted into a direct relationship by considering the one-minusactivity values. In order to easily identify the activity ranges, two groups with 5 values were implemented; the micro-molar concentrations corresponding to APOD values from 0 to 4, and the nano-molar concentrations corresponding to APOD values from 5 to 9. First, using PubChem, the drug name was entered into the search field (PubChem). After browsing through the output, the relevant drug was selected. The default drug data available in the 2D Structure Data Format (SDF) was downloaded to the local computer. The chemical-data file contains a large number of associated data. The only associated data needed for analysis are: pubchem_molecular_weight, pubchem_cactvs_hbond_acceptor, pubchem_cactvs_hbond_donor, pubchem_xlogp3, and pubchem_cactvs_tpsa. The pubchem_compound_cid, a unique drug ID J o u r n a l P r e -p r o o f was available. Since the name of the drug was not available in the associated data, pubchem_common_name was added and the drug name was entered manually by editing the SDF text file and saved; pubchem_bioactivity was also added. Even though, it was not required for the pharmacokinetic analysis, if the experimental information for bioactivity was available in the "Biological Test Results" section of PubChem entry or literature, it was entered as well. This was repeated for all the necessary drugs for the study. Datasets were created after excluding duplicates, synonyms, and monoclonal antibodies such as bevacizumab, eculizumab, mavrilimumab, meplazumab, sarilumab, siltuximab, and tocilizumab. For a specific dataset, text searches with the drug name in PubChem was first performed. Then a list of PubChem identifiers (CIDs) was used as input to PubChem search to view and download records as a SDF. All the drug files were stored together in a sub directory where the program resided. Using APOD, a specific drug or dataset was selected and submit button was clicked. The primary outcome were the physicochemical-pharmacokinetic fingerprint and drug safety profile, numerical-APOD (N-APOD) and APOD score, respectively. The individual values in the N-APOD were interpreted based on the PubChem entries or scientific literature if available. For withdrawn drugs, the ratio of the number of weak drugs predicted to the number of drugs in a specific data set multiplied by 100. Similarly, for approved drugs, the ratio of the J o u r n a l P r e -p r o o f number of strong/moderate drugs predicted to the number of drugs in a specific data set multiplied by 100. J o u r n a l P r e -p r o o f The numerical APOD (N-APOD) included individual physicochemical and pharmacokinetic indicators. The independently determined APOD score was the sole indicator of drug safety. Based on the APOD score, a drug was classified into one of three categories: strong, moderate and weak. For a drug with a strong or moderate safety profile, the APOD value associated with the benefit (first number) was greater than the APOD value associated with the risk (second number). When the difference between the two indicators was large (greater than 4), the drug was categorized as strong; and when the difference was small (less than or equal to 4), it was categorized as moderate. On the contrary, for a drug with a weak safety profile, the risk indicator (second number) was greater than the benefit indicator (first number) in the APOD score, irrespective of the difference between the two indicators. A total of 3556 compounds from PubChem (https://pubchem.ncbi.nlm.nih.gov/) comprising of various unbiased datasets were analyzed based on different criteria such as drugs withdrawn, toxicity, approved, and clinical trials (Table 1) . For the recalled drugs shown in Figure 1 , such as nefazodone and troglitazone, it was evident that the predicted APOD values (indicated by arrows) for absorption (A), distribution (D), and excretion (E) were smaller than those for toxicity (T). For these drugs, the APOD score (of 1-8) indicates that it was alarmingly risky. Justly, these drugs were withdrawn due to the risk of hepatotoxicity. This pattern of the benefit indicator dwarfed by the risk indicator was mostly consistent for the recalled drugs, categorized under weak safety profile in Table 2 . The reasons to withdraw the drug from the market for safety reasons were reported to be stroke, liver toxicity, fatal allergic reaction, hemolytic anemia, flank pain syndrome, fatal arrhythmia, kidney failure, increased deaths, heart value disease, severe constipation, breathing difficulty, birth defects, and skin disease (CDER, 2004). Besides, manufacturers or distributors usually implement voluntary recalls in order to carry out their responsibilities to protect the public health when they need to remove a marketed drug product that presents a risk of injury to consumers or to correct a defective drug product (CDER, 2004) . Out of 24 historic FDA withdrawn drugs, only 4 drugsazaribine, grepafloxacin, flosequinan, and temafloxacin -indicated false positives (categorized under strong/moderate safety profile in Table 2 ). This indicates that the APOD method had a success rate of 83.33%. Interestingly, the false positives for the 4 withdrawn drugs could be due to that fact they seem to be non-ADMET related issues. Azaribine is a prodrug, a biologically inactive compound which must be converted into the pharmacologically active agent (that would usually have issues such as poor aqueous solubility, instability, and other adverse effects or toxicities) by metabolic transformation (Wu, 2009 ). The most common adverse events with grepafloxacin were J o u r n a l P r e -p r o o f gastrointestinal, such as nausea, vomiting, and diarrhea (Lode, Vogel, & Elies, 1999) . Flosequinan increased the heart rate and neurohormonal activation. (Packer et al., 2017) . Temafloxacin was withdrawn due to immune hemolytic anemia (Blum, Graham, & McCloskey, 1994) . Guengerich has listed withdrawn drugs due to hepatotoxicity (Guengerich, 2011 Drugs are a common source of acute kidney injury, also known as nephrotoxicity. Naughton has published the list of drugs withdrawn due to nephrotoxicity (Naughton, 2008) . These 10 FDA withdrawn drugs have the APOD success rate of 80.0%. Cytochrome P450 enzymes are essential for the metabolism of many medications. These enzymes can be inhibited or induced by drugs, resulting in clinically significant drug-drug interactions that can cause unanticipated adverse reactions or therapeutic failures (Lynch and Price, 2007) . All 40 drugs on the list of strong cytochrome P450 CYP3A4 inhibitors (Table 4) For the FDA approved drugs (Drugs@FDA) shown in Figure 2 , such as tylenol and cipro, the trends were opposite. For these drugs, the APOD score of 6-2 indicates that these drugs had a moderate safety profile. This pattern of the benefit indicator outweighing the risk indicator was consistent for the FDA approved drugs (partial list) shown in Table 5 . (Gayvert et al., 2016) . If we consider human disease as an imbalance in the system with increased or decreased effects, then we need every tool in the arsenal of approved drugs to correct or cure by inhibiting or inducing changes to bring the system back into balance. After all, the ultimate goal in finding a cure to a disease is maximizing benefits with minimal or no adverse effects. Naturally, focusing more on the analysis of withdrawn drugs is justified here. The APOD approach demonstrates remarkable sensitivity as well. As indicated in the Table 5 , penicillin G, which was a commonly used penicillin derivative (sodium or potassium salts) on the market, had an APOD score of 6-2, whereas the parent, penicillin, had an APOD score of 5-3, matching with comparatively lower benefit and higher risk. Figure 3C . The patterns and consistencies between the experimental evidence and the independent APOD results are striking. The improvement in the safety profile of the widely used penicillin G is due to the potassium salt penicillin derivative. The progression of the series of cancer drugs from PI-103 to clinical development candidate GDC-0941 is due to the shift in the safety profile favorably ( Figure 3C ) by increasing the polar surface area through rearrangement of the oxygen atoms in the structure and addition of hydrogen-bond acceptors while staying within the ideal property boundaries. This implies that the existing effective drugs with weak safety profile can be modified into effective drugs with moderate/strong profiles through the insights from the APOD fingerprint and difference APOD, even before the molecule is synthesized. if we consider the drug profile spectrum going from strong to weak, evidently the APOD values associated with absorption, distribution and excretion decrease whereas the APOD value associated with toxicity increases (Figure 4) . By sorting the APOD score in the descending order for benefits and ascending order for risks, it is possible to prioritize drugs on the basis of their APOD score (Table 7 ). A representative in the category of strong safety profiles is molnupiravir which includes avigan, aviptadil, azvudine, baricitinib, deferoxamine, emtricitabine, galidesivir, riamilovir, ribavirin, and tenofovir with their APOD scores ranging from 8-1 to 7-2. Molnupiravir, previously EIDD-2801 or MK-4482, is a broad-spectrum antiviral currently in phase III clinical trials. Recently, Wahl and others have shown that molnupiravir dramatically inhibited human coronavirus (SARS-CoV-2) replication in vivo and thus has significant potential for the prevention and treatment of COVID-19 (Wahl et al., 2021) . Their results indicate that molnupiravir did not cause significant mitochondrial toxicity. Molnupiravir has an APOD score of 7-1. Based on its N-APOD, the APOD values associated with absorption, distribution and toxicity were 6, 9 and 0, respectively. With improved oral bioavailability in nonhuman primates, it is hydrolyzed in vivo, and distributes into tissues where it becomes the active 5'-triphosphate form (PubChem). For tenofovir, based on its N-APOD, the APOD values associated with metabolism and toxicity were 4 and 0, respectively. Tenofovir has been shown to be highly effective in patients that have never had an antiretroviral therapy and it seemed to have lower toxicity than other antivirals such as stavudine. In phase 3 clinical trials, tenofovir presented an efficacy profile similar to efavirenz in treatment-naive HIV patients. In hepatitis B infected patients, after one year of tenofovir treatment, the viral DNA levels were undetectable. Tenofovir presents minimal metabolic processing and it does not appear to be a significant cause of drug induced liver injury (PubChem). A representative in the category of moderate safety profiles is dexamethasone which includes AT-527, alpha lipoic acid, anakinra, apremilast, camostat, colchicine, dexamethasone, methylprednisolone, oseltamivir, remdesivir, suramin, and thalidomide with their APOD scores ranging from 6-2 to 5-4. For remdesivir, based on its N-APOD, the APOD values associated with metabolism and toxicity were 7 and 1, respectively. In a study, 537 patients hospitalized for severe COVID-19 who were treated intravenously with compassionate-use remdesivir, clinical improvement was observed and appears to have a favorable clinical safety profile (Grein et al., 2020) . However, it has also been reported that remdesivir was not suitable for oral delivery as its poor hepatic stability, because it was extensively metabolized, would likely result in almost complete firstpass clearance; there were no liver changes in rats or monkeys; remdesivir was non genotoxic in a standard battery of in vitro and in vivo studies (Grein et al., 2020) . J o u r n a l P r e -p r o o f A representative in the category of weak safety profiles is lopinavir which includes abivertinib, acalabrutinib, arbidol, atorvastatin, atovaquone, azithromycin, baloxavir marboxil, bemcentinib, chloroquine, cholecalciferol, ciclesonide, clopidogrel, cobicistat, danoprevir, darunavir, dihydroartemisinine, ebastine, fingolimod, glucocorticoid, glycyrrhizinate, hydroxychloroquine, ifenprodil, imatinib, leflunomide, levamisole, losartan, naproxen, nintedanib, omeprazole, piperaquine, pirfenidone, prasugrel, RBT-9, ritonavir, rivaroxaban, ruxolitinib, selinexor, simvastatin, tetrandrine, tofacitinib, tradipitant, triazavirin, valsartan, and vidofludimus with their APOD scores ranging from 4-4 to 0-8. Liponavir is an antiretroviral protease inhibitor. Based on its N-APOD, the APOD values associated with absorption, metabolism and toxicity were 1, 7 and 5, respectively. Lopinavir has exceptionally low oral bioavailability, undergoes extensive oxidative metabolism by hepatic cytochrome P450, and can cause serious adverse effects (PubChem). Researchers reported that liponavir/ritonavir caused significant liver damage (Cao et al., 2020) . In this study, a mathematical model was used to investigate whether the physicochemical-pharmacokinetic fingerprint was associated with the drug safety profile for investigational treatments of COVID-19. The model was specific for oral formulations of a drug and relies on the best-known scientific data available on a large number of property-safety relationships to predict the pharmacokinetic outcome of a drug. The APOD method encapsulates these associations in a fingerprint of drug and the APOD score reveals the drug safety. For a variety of 417 recalled drugs for failing to meet safety standards demonstrated that the physicochemical-pharmacokinetic fingerprint was associated with the drug safety profile, with an average success rate of 78%. The classification of large number of investigational treatments J o u r n a l P r e -p r o o f of COVID-19 into three categories of drug safety profiles, namely strong, moderate, weak, was also achieved. World Health Organization (WHO) had announced a large global trial called Solidarity for the 6 ( Table 7) promising coronavirus drugs to find out whether they could treat infections with the new coronavirus for the dangerous respiratory disease. As presented here, remdesivir had a moderate safety profile; hydroxychloroquine, chloroquine, and liponavir/ritonavir had a weak safety profile. These safety profiles were consistent with the observations reported (Cao et al., 2020) . FDA approved drugs comprise of safe as well as weak drugs. Drugs with weak safety profiles cannot be repurposed. The safety is an inherent characteristic of the drug, arising purely from the chemical property. Hence, the APOD score is independent of the disease target. Drugs with strong and moderate safety profiles can be repurposed for other disease targets. High throughput screening (HTS) involves the screening of a large compound library directly against the drug target using a cell-based essay. Currently, all compounds are screened for all diseases in the early stages of drug discovery. Due to the iterative nature of the drug discovery process, this has a risk of starting with a set of unsafe drugs with weak safety profiles, thereby wasting precious time and resources. In future, it is recommended that the large compound library databases are screened using the APOD method and a small subset are created for initial screening against a new disease target. This sets a path with much greater success of drug for human therapeutics use. J o u r n a l P r e -p r o o f As demonstrated, the APOD methodology predicts cardiotoxicity, hepatotoxicity, and nephrotoxicity with remarkable successes. However, the limitation of APOD seems to be in predicting adverse effects arising from thromboembolism, hypoglycaemia, skin reactions, hypersensitivity reaction, heartburn, lactic acidosis, birth defects, severe constipation, diarrhea, neurohormonal activation, immune hemolytic anemia, and breathing difficulty. J o u r n a l P r e -p r o o f CONCLUSION In this analysis, the physicochemical-pharmacokinetic fingerprint was associated with the drug safety profile of withdrawn, approved and COVID-19 related drugs. The pharmacokinetic modeling for drug safety was solely focused on the oral formulation of a drug due to the global aspiration of a durable, cost-effective, and easy to administer treatment that can reach everyone including those lacking facilities and resources. The APOD method provides an advancement in the field of drug discovery by unifying all properties in the fingerprint, making any property comparisons possible with uniform representation, and integrating pharmacokinetic insights into drug design. APOD is sensitive and easy to use. The strengths of APOD are interpretability and repeatability of results. APOD provides biological insight not only into absorption, distribution, metabolism, excretion, and toxicity of a drug but also its exact placement in the spectrum of drug safety profiles. The prioritization of drugs based on their safety profiles allows for the acceleration of the pace of drug discovery. The five physicochemical properties needed for drug safety analysis are already available as online compound data, making the instantaneous extraction of the pharmacokinetic properties to produce an APOD score, which is independent of the disease target, highly convenient. Future work will focus on making the APOD score available routinely for all drugs that are part of public databases. J o u r n a l P r e -p r o o f For interactive session, the resource is freely available at: https://apodvision.com/apod/apod.pl. However, there are restrictions for batch processing using SDF. ADMET -turning chemicals into drugs A primer on effectiveness and efficacy trials The penicillins Report to the Nation, the U.S. Department of Health and Human Services, Food and Drug Administration A New Classification of Prodrugs: Regulatory Perspectives Grepafloxacin: a review of its safety profile based on clinical trials and postmarketing surveillance Effects of Flosequinan on the Morbidity and Mortality of Patients With Severe Chronic Heart Failure: Primary Results of the PROFILE Trial After 24 Years Temafloxacin syndrome: review of 95 cases Biological properties of potent inhibitors of class I phosphatidylinositide 3-kinases: from PI-103 through PI-540, PI-620 to the oral agent GDC-0941 Abbreviated Profile of Drugs (A-POD): A unique numerical and graphical representation for compound properties and its use in ADMET prediction Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings In silico ADME prediction: data, models, facts and myths Medicinal chemical properties of successful central nervous system drugs Experimental and computational screening models for prediction of intestinal drug absorption Property-based design: optimization of drug absorption and pharmacokinetics ADMET Property Prediction: The State of the Art and Current Challenges A Data-Driven Approach to Predicting Successes and Failures of Clinical Trials An integrative machine learning approach for prediction of toxicity-related drug safety Applications of machine learning in drug discovery and development Machine Learning Methods in Drug Discovery SwissADME: a free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules Computational Approaches in Preclinical Studies on Drug Discovery and Development Withdrawing Drugs in the U.S. Versus Other Countries Withdrawn medicines included in the essential medicines lists of 136 countries DILIrank: the largest reference drug list ranked by the risk for developing drug-induced liver injury in humans Human Induced Pluripotent Stem Cell-Derived Cardiomyocytes Afford New Opportunities in Inherited Cardiovascular Disease Modeling Mechanisms of drug toxicity and relevance to pharmaceutical development. Drug metabolism and pharmacokinetics Drug-induced nephrotoxicity The effect of cytochrome P450 metabolism on drug response, interactions, and adverse effects SARS-CoV-2 infection is effectively treated and prevented by EIDD-2801 Advance online publication Compassionate Use of Remdesivir for Patients with Severe Covid-19 A Trial of Lopinavir-Ritonavir in Adults Hospitalized with Severe Covid-19