Background - Measurements and predictions of aqueous solubility (S), apparent cell permeability (Papp), unbound microsomal intrinsic clearance (CLint,u), unbound fraction in plasma (fu), log D, Lipinskis Rule of 5 (Ro5) and BBB+/BBB- (blood-brain-barrier) are commonly used in early drug discovery to evaluate whether compounds are likely to have adequate ADME/PK in humans. The main objective was to evaluate the validity and proposed thresholds for commonly used in silico to in vitro (with particular interest in the ADMET Predictor software) and in vitro to in vivo human ADME/PK prediction methods, and physicochemical estimates and rules of thumb. A secondary aim was to compare validity and thresholds to that for the prediction software ANDROMEDA. Methods - Data were collected from literature and own studies. Main measures of validity were Q2 (true predictive accuracy; in silico models), R2 (correlation coefficient; in vitro models), Q2 x R2 (for translation from in silico to in vivo via in vitro), skewness, range, % correct predicted class and clinical relevance. Results and Discussion - Poor accuracies (Q2 x R2=0.05, 0.05, 0.36 and 0.45; ca 0 at low to moderate, decision-critical levels) and class predictions, limited ranges (not covering low to moderate estimates), systematic errors (often considerable overprediction at low values), poor clinical relevance and inadequate thresholds were found. Predictive accuracy was mainly lost in the translation from in vitro to in vivo. Log D and Ro5 are poor predictors of oral bioavailability and half-life. Ro5 produced 63 % false negatives for prediction of poor/good oral absorption. The overall mean Q2 for ANDROMEDA was 3 times higher (0.57 vs 0.2). Advantages with ANDROMEDA compared to in vitro data-based in silico models include wider application domain (high resolution at low values), extrahepatic elimination models, minimal skewness, clinical relevance, balanced thresholds, and compound- and parameter-unique confidence intervals. ANDROMEDA successfully predicted the human clinical ADME/PK for all small drugs marketed in 2021, while the in silico to in vitro to in vivo approach was out of reach for all. Conclusion - The validity of investigated methodologies (including ADMET Predictor) and thresholds were overall very low. Unless both predicted S, permeability and fu are high and CLint is moderate (an overall criterium not met for investigated modern small drugs) one is more or less lost in the translation, and will jeopardize compound selection and optimization. This clearly shows the need for better and thoroughly validated prediction models and software. Marked improvements in accuracy, range, balance and clinical relevance were achieved with ANDROMEDA, which predicts human clinical ADME/PK directly from chemical structures and has undergone extensive validation. ### Competing Interest Statement Urban Fagerholm and Sven Hellberg declare shares in Prosilico AB, a Swedish company that develops solutions for human clinical ADME/PK predictions.

Background - The influential Lipinskis Rule of 5 (Ro5) describes molecular properties important for oral absorption in humans. According to Ro5, poor absorption is more likely when 2 or more of its criteria (molecular weight (MW) above 500 g/mol, calculated octanol-water partition coefficient (clog P) above 5, above 5 hydrogen bond donors (HBD) and above 10 hydrogen bond acceptors (HBA)) are violated. Earlier evaluations have shown that many drugs are sufficiently well absorbed into the systemic circulation despite many Ro5-violations. No evaluation of Ro5 vs fraction absorbed (fa) has, however, been done. Methods - Datasets of orally administered drugs violating and not violating Ro5 and with available human clinical fa-values were assembled, and contrasted to machine learning based predictions using the ANDROMEDA prediction software having a major MW-domain of 150-750 g/mol. Results - 129 Ro5-violent compounds (29 with MW above 1000 g/mol) were found, 59 of which had fa-values (42 % mean fa). 34 % and 66 % of compounds were predicted as having fa below 10 % and above 10-30 % respectively, which was in good agreement with measured fa of 37 % and 63 %. The fa for all compounds with fa above 5 % and above 10 % were correctly predicted. For compounds with fa above 30 %, 81 % were predicted to have a fa above 30 %, but none were predicted to have a fa below 10 %. The Q2 for predicted vs observed fa was 0.64. For a set of 77 compounds without Ro5 violation (80 % mean fa), all compounds were correctly predicted to have a fa below or above 30 % (Q2=0.56). Among these are compounds with poor uptake (below 1 % to 7 %). Conclusion - We show that machine learning based predictions of fa are superior to Ro5 for assessing oral absorption obstacles in humans. Too strict reliance on Ro5 may hence constitute a risk. ANDROMEDA predicts fa well, easily and quickly, and also differentiates well between poor and adequate oral uptake for compounds violating and not-violating Ro5. This makes it a valid and useful tool capable of predicting oral absorption in humans with good accuracy and replacing Ro5 for oral absorption assessments. ### Competing Interest Statement Urban Fagerholm, Sven Hellberg, Morgan Ekmefjord and Ola Spjuth declare shares in Prosilico AB, a Swedish company that develops solutions for human clinical ADME/PK predictions. Ola Spjuth declares shares in Aros Bio AB, a company developing the CPSign software.

There is an ongoing aim to replace animal and in vitro laboratory models with in silico methods. Such replacement requires the successful validation and comparably good performance of the alternative methods. We have developed an in silico prediction system for human clinical pharmacokinetics, based on machine learning, conformal prediction and a new physiologically-based pharmacokinetic model, i.e. ANDROMEDA. The objectives of this study were: a) to evaluate how well ANDROMEDA predicts the human clinical pharmacokinetics of a previously proposed benchmarking data set comprising 24 physicochemically diverse drugs and 28 small drug molecules new to the market in 2021; b) to compare its predictive performance with that of laboratory methods; and c) to investigate and describe the pharmacokinetic characteristics of the modern drugs. Median and maximum prediction errors for the selected major parameters were ca 1.2 to 2.5-fold and 16-fold for both data sets, respectively. Prediction accuracy was on par with, or better than, the best laboratory-based prediction methods (superior performance for a vast majority of the comparisons), and the prediction range was considerably broader. The modern drugs have higher average molecular weight than those in the benchmarking set from 15 years earlier ( ca 200 g/mol higher), and were predicted to (generally) have relatively complex pharmacokinetics, including permeability and dissolution limitations and significant renal, biliary and/or gut-wall elimination. In conclusion, the results were overall better than those obtained with laboratory methods, and thus serve to further validate the ANDROMEDA in silico system for the prediction of human clinical pharmacokinetics of modern and physicochemically diverse drugs.

ABSTRACT The ANDROMEDA toolkit for prediction of human clinical pharmacokinetics, based on machine learning, conformal prediction and a new physiologically-based pharmacokinetic model, was used to predict and characterize the human clinical pharmacokinetics of 12 small anticancer drugs marketed in 2021 and 2022 (molecular weight 355 to 1326 g/mol). The study is part of a series of software validations. A majority of clinical pharmacokinetic data was missing. ANDROMEDA successfully filled this gap. Most drugs were predicted/measured to have relatively complex pharmacokinetics, with limited passive permeability+efflux, high degree of plasma protein binding, significant gut-wall elimination and food interaction, biliary excretion and/or limited dissolution potential. Median, mean and maximum prediction errors for steady state volume of distribution, unbound fraction in plasma, blood-to-plasma concentration ratio, hepatic, renal and total clearance, fraction absorbed, oral bioavailability, half-life and degree of food interaction were 1.6-, 2.4- and 17-fold, respectively. Less than 3-fold errors were found for 78 % of predictions. Results are consistent with those obtained in previous validation studies and are better than with the best laboratory-based prediction methods, which validates ANDROMEDA for predictions of human clinical pharmacokinetics of modern small anticancer drugs with multi-mechanistical and challenging pharmacokinetics.

ABSTRACT The ANDROMEDA software, based on machine learning, conformal prediction and a new physiologically-based pharmacokinetic model, was used to predict and characterize the human clinical pharmacokinetics of 30 selected modern small antibiotic compounds (investigational and marketed drugs). A majority of clinical pharmacokinetic data was missing. ANDROMEDA successfully filled this gap. Most antibiotics were predicted and measured to have limited permeability, good metabolic stability and multiple elimination pathways. According to predictions, most of the antibiotics are mainly eliminated renally and biliary and every other antibiotic is mainly eliminated via the renal route. Mean prediction errors for steady state volume of distribution, unbound fraction in plasma, renal and total clearance, oral clearance, fraction absorbed, fraction excreted renally, oral bioavailability and half-life were 1.3- to 2.3-fold. The overall median and maximum prediction errors were 1.5- and 4.8-fold, respectively, and 92 % of predictions had <3-fold error. Results are consistent with those obtained in previous validation studies and are better than with the best laboratory-based prediction methods, which validates ANDROMEDA for predictions of human clinical pharmacokinetics of modern antibiotic drugs, which to a great extent demonstrate pharmacokinetic characteristics challenging for laboratory methods (metabolic stability, limited permeability, efflux and multiple elimination pathways). Advantages with ANDROMEDA include that results are produced without the use of animals and cells and that predictions and decision-making can be done already at the design stage.

A bstract Introduction In vitro measurements and predictions of human clinical pharmacokinetics (PK) are sometimes hindered and made impossible due to factors such as extensive binding to materials, low methodological sensitivity and large variability. Methods The objective was to find compounds out of reach for in vitro PK-methods and (if possible) predict corresponding human clinical estimates using the ANDROMEDA by Prosilico software. In vitro methods selected for the investigation were human microsomes and hepatocytes for measuring and predicting intrinsic hepatic metabolic clearance (CL int ), Caco-2 and Ralph Russ canine kidney cells (RRCK) cells for measuring apparent intestinal permeability (P app ) for prediction of fraction absorbed (f a ), plasma for measurement and estimation of unbound fraction (f u ), and water and buffers for measuring solubility (S) for prediction of in vivo dissolution potential (f diss ). Results and Conclusion As many as 329 non-quantifiable in vitro PK-measurements for 300 compounds were found in the literature: 191 for CL int , 101 for P app , 11 for f u and 26 for S. ANDROMEDA was successful in predicting all corresponding clinical PK-estimates for the selection of compounds with non-quantifiable in vitro PK, and predicted estimates (1.6-fold median prediction error; n=159) were generally in line with observed in vivo data and results/problems at in vitro laboratories. Thus, ANDROMEDA is applicable for predicting human clinical PK for compounds out of reach for laboratory methods.

In vitro-in vivo prediction results for hepatic metabolic clearance (CLH) and intrinsic CLH (CLint) vary widely among studies. Reasons are not fully investigated and understood. The possibility to select favorable reference data for in vivo CLH and CLint and unbound fraction in plasma (fu) is among possible explanations. The main objective was to investigate how reference data selection influences log in vitro and in vivo CLint-correlations (r2). Another aim was to make a head-to-head comparison vs an in silico prediction method. Human hepatocyte CLint-data for 15 compounds from two studies were selected. These were correlated to in vivo CLint estimated using different reported CLH- and fu-estimates. Depending on the choice of reference data, r2 from two studies were 0.07 to 0.86 and 0.06 to 0.79. When using average reference estimates a r2 of 0.62 was achieved. Inclusion of two outliers in one of the studies resulted in a r2 of 0.38, which was lower than the predictive accuracy (q2) for the in silico method (0.48). In conclusion, the selection of reference data appears to play a major role for demonstrated predictions and the in silico method showed higher accuracy and wider range than hepatocytes for human in vivo CLint-predictions.

A bstract Background Passive blood-brain barrier permeability (BBB P e ), fraction bound to brain tissue (f b,brain ) and efflux by transport proteins MDR-1 and BCRP are essential determinants for the brain uptake and disposition of drugs. Methods The main objective of the study was to use the software ANDROMEDA by Prosilico to predict passive BBB P e - and f b,brain -classes and MDR-1- and BCRP-specificities for various classes of antidepressants and for CNS-active small drugs marketed during 2020 and 2021, and then to position them according to a new 2-dimensional Brainavailability-Matrix (8 passive BBB P e x 4 f b,brain classes, where class 11 has highest and 84 lowest values/brainavailability). Predicted estimates were used, except for cases where measured values were available. Results and Conclusion Results for 53 drugs show that adequate CNS uptake and disposition are achieved for compounds placed in the zones for low, moderate and high brainavailability, despite efflux. They also show that high brainavailability and efflux are common for CNS-active drugs and that modern CNS-active drugs generally have lower brainavailability than older antidepressive drugs. Furthermore, they demonstrate that ANDROMEDA by Prosilico and the new Brainavailability-Matrix are applicable for prediction, optimization and positioning of CNS uptake and disposition of drugs and drug candidates in man.

AbstractBackgroundIt is important that pharmacokinetic (PK) prediction methods are validated, and also for compounds with varying physicochemical properties, molecular weights and PK characteristics.MethodsThe objective was to investigate how well the ANDROMEDA by Prosilico software predicts the clinical PK of four compounds of natural origin and with PK obstacles, not yet fully characterized PK, and/or inaccurate lab method-based predictions - colistin (negligible absorption, good metabolic stability, significant excretion), curucumin (low solubility, apparently poor bioavailability), UCN-01 (extremely high degree of plasma protein binding, metabolic stability, long half-life and poor PK prediction) and voclosporin (poorly understood PK).ResultsAll categorial predictions except one were correct, and the median prediction error was 2.5-fold. Largest prediction errors were found for the unbound fraction in plasma (>24-fold), clearance (178-fold) and half-life (90-fold) of UCN-01. Corresponding errors for clearance and half-life obtained with allometry were greater, 5800- and 145-fold, respectively. Extremely high affinity for alpha1-acid glycoprotein could explain these large prediction errors for this compound. A substantial amount of data and knowledge was added with the predictions.ConclusionDespite challenging compounds and PK, predictions were comparably good. The results further validated ANDROMEDA by Prosilico for human clinical PK-predictions.

AbstractIntroductionSome prodrugs are developed in order to improve gastrointestinal absorption properties such as permeability and solubility/dissolution. Prediction of the uptake of prodrugs and their drugs is challening for reasons including gastrointestinal hydrolysis and active transport.Objective and MethodologyThe objective was to use the ANDROMEDA software by Prosilico to predict absorption characteristics - passive fraction absorbed (fa,passive), dose-adjusted dissolution potential (fdiss) and total fa(fa) - of prodrugs and their drugs (including drugs and their active metabolites), and to evaluate how they differ between prodrugs and drugs and the predictive accuracy of the software.Results70 prodrug-drug pairs were found and selected for the study. The mean predicted fa,passiveand fdissfor the prodrugs were 0.74 and 0.94, respectively. Corresponding estimates for the drugs were 0.72 and 0.98, respectively. For non-hydrolyzed prodrugs, the median relative and absolute prediction errors for fawere 1.17-fold and 0.08, respectively. Corresponding values for drugs were 1.11-fold and 0.07, respectively. The correlation between predicted and observed fafor non-hydrolyzed ester prodrugs and drugs combined (predictive accuracy) was 0.6.ConclusionProdrugs and drugs had similar average predicted fa,passiveand fdiss, and most had or were predicted to have at least 50 % fa. The fafor about 1/3 of non-hydrolyzed prodrugs was higher than for corresponding drugs, showing successful prodrug design. Adequate prediction accuracy validates ANDROMEDA for prediction of prodrug and drug absorption in man.

A bstract Introduction The ANDROMEDA software by Prosilico has previously been successfully applied and validated for predictions of absorption characteristics of small drugs in man. The influence of fat food on the gastrointestinal uptake and systemic exposure of drugs have, however, not yet been evaluated with the software. Objective and Methodology The main objective was to use ANDROMEDA to predict area under the plasma concentration-time curve ratios in the fed (fat food) and fasted states (AUC fed /AUC fast ) for small drugs (including those marketed in 2021) and compare results with corresponding measured clinical estimates. Actual dose sizes were considered. Another objective was to compare the performance of ANDROMEDA vs physiologically based pharmacokinetic (PBPK) modelling and simulations by The Food Effect PBPK IQ Working Group. PBPK results generated using Simcyp and GastroPlus software were based on various physicochemical, in vitro and in vivo data and a decision tree for model verification and optimization. Results and Discussion 63 drugs, including 17 new drugs, with observed AUC fed /AUC fast between 0.2 and 5.5 were found and used for this evaluation. Predicted AUC fed /AUC fast had mean and maximum errors of 1.5- and 4.1-fold, respectively, and the predictive accuracy (correlation between predicted and observed AUC fed /AUC fast ; Q 2 ) was 0.3. 14 % of predictions had >2-fold error. For 72 % of drugs, food interaction class was correctly predicted. The level of predictive accuracy was overall similar to results obtained with PBPK modelling and simulations, however, with lower maximum error and higher compound coverage. With PBPK models, maximum simulation error was 7.7-fold and 3 highly lipophilic compounds were not possible to simulate. Conclusion The results validate ANDROMEDA for prediction of fat food-drug interaction size for small drugs in man. Major advantages with the methodology include that prediction results are produced directly from molecular structure and oral dose and are similar to PBPK-simulation results obtained using in vitro and clinical data. Furthermore, ANDROMEDA showed lower maximum errors and wider compound range.

ABSTRACT Introduction Conformal prediction (CP) methodology sits on top of machine learning methods and produces prediction confidence intervals that depend on how “strange” (non-conforming) test compounds are compared to training set compounds. CP has previously been successfully applied for prediction of steady-state volume of distribution (V ss ) in humans, with 69 % of observations within the prediction interval at a 70 % confidence level. We have developed CP models for a variety of human pharmacokinetic (PK) parameters and validated their predictive accuracy (predicted vs observed estimates), but not validated prediction confidence intervals for them. The main objective of this study was to predict 70 % confidence intervals for V ss , unbound fraction in plasma (f u ), intrinsic metabolic clearance (CL int ), fraction absorbed passively (f a,passive ) and maximum fraction dissolved (f diss ) for a variety of compounds in man and investigate the consistency between prediction intervals and observed/measured values. Methodology CP models featured in the ANDROMEDA software by Prosilico were used for prediction of 70 % confidence intervals of V ss , f u , CL int , f a,passive and f diss for compounds from different chemical classes and with broad physicochemical variety and for small drugs marketed in 2021. Results 70 % prediction confidence intervals for 217, 117, 117, 89 and 89 compounds were produced for V ss , f u , CL int , f a,passive and f diss , respectively. 78 % (expected 70 %) of observed data were within 70 % confidence intervals for the parameters. 70 % of predictions of V ss , f u , CL int f a,passive and f diss are expected to have errors of maximally 2-, 4- and 6-fold and 7 and 12 %, respectively, which is in line with prediction errors. These findings validate the CP methodology. Conclusion In conclusion, the results further validate CP models and confidence intervals of ANDROMEDA for prediction of human PK.

Pharmacokinetic/toxicokinetic (PK/TK) information for chemicals in humans is generally lacking. Here we applied machine learning, conformal prediction and a new physiologically-based PK/TK model for prediction of the human PK/TK of 65 chemicals from different classes, including carcinogens, food constituents and preservatives, vitamins, sweeteners, dyes and colours, pesticides, alternative medicines, flame retardants, psychoactive drugs, dioxins, poisons, UV-absorbents, surfactants, solvents and cosmetics.About 80% of the main human PK/TK (fraction absorbed, oral bioavailability, half-life, unbound fraction in plasma, clearance, volume of distribution, fraction excreted) for the selected chemicals was missing in the literature. This information was now added (from in silico predictions). Median and mean prediction errors for these parameters were 1.3- to 2.7-fold and 1.4- to 4.8-fold, respectively. In total, 59 and 86% of predictions had errors <2- and <5-fold, respectively. Predicted and observed PK/TK for the chemicals was generally within the range for pharmaceutical drugs.The results validated the new integrated system for prediction of the human PK/TK for different chemicals and added important missing information. No general difference in PK/TK-characteristics was found between the selected chemicals and pharmaceutical drugs.

The gastrointestinal uptake of macrocyclic compounds is not fully understood. Here we applied our previously validated integrated system based on machine learning and conformal prediction to predict the passive fraction absorbed (fa), maximum fraction dissolved (fdiss), substrate specificities for major efflux transporters and total fraction absorbed (fa,tot) for a selected set of designed macrocyclic compounds (n = 37; MW 407-889 g/mol) and macrocyclic drugs (n = 16; MW 734-1203 g/mole) in vivo in man. Major aims were to increase the understanding of oral absorption of macrocycles and further validate our methodology. We predicted designed macrocycles to have high fa and low to high fdiss and fa,tot, and average estimates were higher than for the larger macrocyclic drugs. With few exceptions, compounds were predicted to be effluxed and well absorbed. A 2-fold median prediction error for fa,tot was achieved for macrocycles (validation set). Advantages with our methodology include that it enables predictions for macrocycles with low permeability, Caco-2 recovery and solubility (BCS IV), and provides prediction intervals and guides optimization of absorption. The understanding of oral absorption of macrocycles was increased and the methodology was validated for prediction of the uptake of macrocycles in man.

ABSTRACTBackgroundPROTACs are comparably large and flexible compounds with limited solubility (S) and permeability (Pe). It is crucial to better understand, predict and optimize their human clinical pharmacokinetics (PK).MethodsThe main objective was to use the ANDROMEDA by Prosilico software to predict the human clinicalin vivodissolution potential (fdiss) and fraction absorbed (fa) of 23 PROTACs at a dose level of 50 mg and to explore whether there is any relationship betweenin vitroS andin silicopredictedin vivofdiss.ResultsIn silicopredictions showed that the PROTACs are effluxed by intestinal transporters and have limited fdiss(34 to 98 %), permeability and fa(13 to 58 %) in man. For some PROTACs this may be a major obstacle and jeopardize the clinical development programs, especially in cases of required high oral dose. A modest relationship betweenin vitroS and predictedin vivofdisswas demonstrated (R2=0.26). Predicted human fa(27 %) and oral bioavailability (20 %) of ARV-110 (a PROTAC with some availablein vivoPK data in rodents and man) were consistent with data obtained in rodents (estimated faapproximately 30-40 %; measured oral bioavailability 27-38 %). Laboratories were unable to quantify S for 7 (30 %) of the PROTACs. In contrast, ANDROMEDA could predict parameters for all.ConclusionANDROMEDA predicted fdissand fafor all the chosen PROTACs and showed limited fdiss, Peand faand dose-dependent fdissand fa. One available example shows promise for the applicability of ANDROMEDA for predicting biopharmaceutics of PROTACsin vivoin man. Weak to modest correlations between S and fdissand a considerable portion of compounds with non-quantifiable S limit the use of S-data to predict the uptake of PROTACs.

Variability of the unbound fraction in plasma (f(u)) between labs, methods and conditions is known to exist. Variability and uncertainty of this parameter influence predictions of the overall pharmacokinetics of drug candidates and might jeopardise safety in early clinical trials. Objectives of this study were to evaluate the variability of human in vitro f(u)-estimates between labs for a range of different drugs, and to develop and validate an in silico f(u)-prediction method and compare the results to the lab variability. A new in silico method with prediction accuracy (Q(2)) of 0.69 for log f(u) was developed. The median and maximum prediction errors were 1.9- and 92-fold, respectively. Corresponding estimates for lab variability (ratio between max and min f(u) for each compound) were 2.0- and 185-fold, respectively. Greater than 10-fold lab variability was found for 14 of 117 selected compounds. Comparisons demonstrate that in silico predictions were about as reliable as lab estimates when these have been generated during different conditions. Results propose that the new validated in silico prediction method is valuable not only for predictions at the drug design stage, but also for reducing uncertainties of f(u)-estimations and improving safety of drug candidates entering the clinical phase.

Volume of distribution at steady state (V-ss) is an important pharmacokinetic endpoint. In this study we apply machine learning and conformal prediction for human V-ss prediction, and make a head-to-head comparison with rat-to-man scaling, allometric scaling and the Rodgers-Lukova method on combined in silico and in vitro data, using a test set of 105 compounds with experimentally observed V-ss. The mean prediction error and % with <2-fold prediction error for our method were 2.4-fold and 64%, respectively. 69% of test compounds had an observed V-ss within the prediction interval at a 70% confidence level. In comparison, 2.2-, 2.9- and 3.1-fold mean errors and 69, 64 and 61% of predictions with <2-fold error was reached with rat-to-man and allometric scaling and Rodgers-Lukova method, respectively. We conclude that our method has theoretically proven validity that was empirically confirmed, and showing predictive accuracy on par with animal models and superior to an alternative widely used in silico-based method. The option for the user to select the level of confidence in predictions offers better guidance on how to optimise V-ss in drug discovery applications.

Quantitative structure-activity relationships (OSAR) constitute empirical analogy models connecting chemical structure and biological activity. The analogy approach to QSAR assume that the factors important in the biological system also are contained in chemical model systems. The development of a QSAR can be divided into subproblems: lj_ to quantify chemical structure in terms of latent variables expressing analogy, 2 _̂ to design test series of compounds, 3u_ to measure biological activity and 4. to construct a mathematical model connecting chemical structure and biological activity. In this thesis it is proposed that many possibly relevant descriptors should be considered simultaneously in order to efficiently capture the unknown factors inherent in the descriptors. The importance of multivariately and multipositionally varied test series is discussed. Multivariate projection methods such as PCA and PLS are shown to be appropriate far QSAR and to closely correspond to the analogy assumption. The multivariate analogy approach is applied to a betaadrenergic agents, b haloalkanes, c halogenated ethyl methyl ethers and d four different families of peptides.

Abnormal tau phosphorylation resulting in detachment of tau from microtubules and aggregation are critical events in neuronal dysfunction, degeneration, and neurofibrillary pathology seen in Alzheimer's disease. Glycogen synthase kinase‐3β (GSK3β) is a key target for drug discovery in the treatment of Alzheimer's disease and related tauopathies because of its potential to abnormally phosphorylate proteins and contribute to synaptic degeneration. We report the discovery of AZD1080, a potent and selective GSK3 inhibitor that demonstrates peripheral target engagement in Phase 1 clinical studies. AZD1080 inhibits tau phosphorylation in cells expressing human tau and in intact rat brain. Interestingly, subchronic but not acute administration with AZD1080 reverses MK‐801‐induced deficits, measured by long‐term potentiation in hippocampal slices and in a cognitive test in mice, suggesting that reversal of synaptic plasticity deficits in dysfunctional systems requires longer term modifications of proteins downstream of GSK3β signaling. The inhibitory pattern on tau phosphorylation reveals a prolonged pharmacodynamic effect predicting less frequent dosing in humans. Consistent with the preclinical data, in multiple ascending dose studies in healthy volunteers, a prolonged suppression of glycogen synthase activity was observed in blood mononuclear cells providing evidence of peripheral target engagement with a selective GSK3 inhibitor in humans.