Advances in machine learning and artificial intelligence have recently extended to the quantitative prediction of drug-drug interaction (DDI). Because DDIs arise from diverse mechanisms and the required level of predictive accuracy varies with both the endpoint and the stage of drug development, evaluating their significance and deciding what is needed demand unusually broad expertise-ranging from fundamental biology all the way to state-of-the-art machine-learning methods. In this review, DMPK scientists with expertise in machine learning survey and critique the most recent literature covering the following DDI categories: Cytochrome P450 (CYP) substrates, CYP competitive and time-dependent inhibition, CYP induction, non-CYP substrates, non-CYP inhibition, transporter substrates, transporter inhibition, and cutting-edge predictive algorithms based on deep learning applied for the task of DDIs. For each category we summarize current in silico methodologies and their performance, and we provide expert opinions on how these tools can be optimally incorporated into contemporary drug-discovery workflows.

The conversion of cholesterol to 24S-hydroxycholesterol (24HC) is a brain-specific reaction catalyzed by cholesterol 24-hydroxylase (CH24H). Soticlestat, a specific inhibitor of CH24H, has the potential to reduce 24HC levels in the brain; 24HC is a neuroactive oxysterol that modulates several receptors and ion channels involved in regulating neural excitability. Soticlestat has been investigated as an additional treatment option for seizures associated with Dravet and Lennox-Gastaut syndromes. Comprehensive preclinical evaluation of pharmacokinetics (PK), target enzyme occupancy (EO), and pharmacodynamics (PDs)-including confirmation of their interrelationships within the brain as the target tissue-is essential for elucidating the mechanism of action of soticlestat. Preclinical data on PK, EO, and PD in rodent models facilitate translational modeling, thereby informing rational human dose selection. A divergence in the temporal profiles of soticlestat concentrations between brain and plasma was observed, with prolonged brain exposure relative to plasma clearance, consistent with the compound's slow dissociation from CH24H. Herein, we characterized the PK, PD, and EO profiles of soticlestat in rodents and demonstrated the relationship between CH24H inhibition and 24HC reduction. Soticlestat demonstrated a delayed-effect model in mice, with time lags between PK and PD as well as between brain and plasma pharmacokinetics. Subsequently, the steady-state PK/PD analysis after subcutaneous infusion in CH24H wild-type [CH24H(+/+)] and knockout [CH24H(-/-)] mice provided comprehensive insights into the model linking the PK, EO, and PD profiles of soticlestat. SIGNIFICANCE STATEMENT: In the present study, a clear relationship between the pharmacokinetics, target enzyme occupancy, and pharmacodynamic markers of cholesterol 24-hydroxylase for soticlestat in rodents. The preclinical pharmacokinetic/enzyme occupancy/pharmacodynamic data for soticlestat provide comprehensive mechanistic and quantitative insights into clinical population models. Importantly, these data emphasize the potential of enzyme occupancy and pharmacodynamics as strategic tools for facilitating translational research in central nervous system drug development.

The conversion of cholesterol to 24S-hydroxycholesterol (24HC) is a brain-specific reaction catalyzed by cholesterol 24-hydroxylase (CH24H). Soticlestat, a specific inhibitor of CH24H, has the potential to reduce 24HC levels in the brain; 24HC is a neuroactive oxysterol that modulates several receptors and ion channels involved in regulating neural excitability. Soticlestat has been investigated as an additional treatment option for seizures associated with Dravet and Lennox-Gastaut syndromes. Comprehensive preclinical evaluation of pharmacokinetics (PK), target enzyme occupancy (EO), and pharmacodynamics (PDs)& horbar;including confirmation of their interrelationships within the brain as the target tissue & horbar;is essential for elucidating the mechanism of action of soticlestat. Preclinical data on PK, EO, and PD in rodent models facilitate translational modeling, thereby informing rational human dose selection. A divergence in the temporal profiles of soticlestat concentrations between brain and plasma was observed, with prolonged brain exposure relative to plasma clearance, consistent with the compound's slow dissociation from CH24H. Herein, we characterized the PK, PD, and EO profiles of soticlestat in rodents and demonstrated the relationship between CH24H inhibition and 24HC reduction. Soticlestat demonstrated a delayed-effect model in mice, with time lags between PK and PD as well as between brain and plasma pharmacokinetics. Subsequently, the steady-state PK/PD analysis after subcutaneous infusion in CH24H wild-type [CH24H (+/+) ] and knockout [CH24H (- /- ) ] mice provided comprehensive insights into the model linking the PK, EO, and PD profiles of soticlestat. Significance Statement: In the present study, a clear relationship between the pharmacokinetics, target enzyme occupancy, and pharmacodynamic markers of cholesterol 24-hydroxylase for soticlestat in rodents. The preclinical pharmacokinetic/enzyme occupancy/pharmacodynamic data for soticlestat provide comprehensive mechanistic and quantitative insights into clinical population models. Importantly, these data emphasize the potential of enzyme occupancy and pharmacodynamics as strategic tools for facilitating translational research in central nervous system drug development. (c) 2026 Takeda Pharmaceutical Company Limited. Published by Elsevier Inc. on behalf of American Society for Pharmacology and Experimental Therapeutics. This is an open access article under the CC BY

Background and aims:Charcot-Marie-Tooth type 1A (CMT1A) is the most common inherited demyelinating peripheral neuropathy caused by duplication of the peripheral myelin protein 22 (PMP22) gene. Although there is currently no approved treatment for CMT1A, reducing PMP22 expression has emerged as a promising therapeutic approach. The PMP22-C3 mouse model is a widely used CMT1A model that carries human PMP22 (hPMP22) and mouse Pmp22 (mPmp22) genes, complicating the relationship between reduced PMP22 levels and the recovery of phenotypes by drug candidates targeting only human PMP22. To address this, we developed humanized C3 mouse lines lacking the mPmp22 gene. Here, we characterized these models to confirm their utility as novel disease models for CMT1A. Methods:Heterozygous (hetero-humanized) and homozygous (homo-humanized) hPMP22 transgenic mice, with an mPmp22 homozygous knockout background, were investigated using biochemical, electrophysiological, histopathological, and behavioral analyses. Results:Homo-humanized mice exhibited abnormal mRNA expression of myelin-related genes, slow nerve conduction velocity, reduced compound muscle action potential, demyelinated peripheral nerves, higher levels of plasma neurofilament light chain, muscle weakness and motor/balance disabilities, alterations in electrical impedance myography, and muscle fiber atrophy. In contrast, the hetero-humanized mice did not display any of the previously described impairments. Interpretation:Homo-humanized mice reflect various aspects of CMT1A characteristics in an hPMP22 gene dosage-dependent manner. This model will help us better understand the relationship between PMP22 reduction levels and the recovery of CMT1A-related phenotypes, contributing to the translation of preclinical findings into clinically relevant human treatments and dosing strategies.

Histone deacetylase 6 (HDAC6) is a crucial target for the development of pharmaceuticals used in the treatment of neurodegenerative disorders. Here, we identified 16a as a candidate of positron emission tomography (PET) tracer for HDAC6 imaging from pyrazole derivatives, which showed strong HDAC6 affinity (Kd = 1.66 nM) and higher accumulation in the brain of wild-type mice than in HDAC6 knockout mice. Following radiolabeling with fluorine-18, PET with [18F]16a exhibited heterogeneous uptake of radioactivity, corresponding to the biological distribution of HDAC6 in the monkey brain. These radioactive distributions were homogeneously diminished by the preadministration of ACY-775, a potent inhibitor of HDAC6, suggesting that radioactive accumulation in PET images could reflect the specific binding of [18F]16a with HDAC6. Thus, [18F]16a is a promising PET tracer for HDAC6 imaging that motivates future clinical research.

Trimethyltin chloride (TMT) is a neurotoxicant that damages the central nervous system (CNS) and triggers neurodegeneration. This study used multi-omic data, including transcriptomics and proteomics of the rat hippocampus, to identify differentially expressed genes and proteins in TMT-induced neurotoxicity over time, related to neuro-axonal damage marked by plasma Neurofilament Light (NfL) levels. Data were collected at 12, 24, 48, 72, and 168 h post-TMT administration. NfL levels surged at 72 and 168 h, confirming neuro-axonal damage. Transcripts of genes in the chemokine signaling pathway (Cxcl10, Cxcl12, Cxcl14, Cxcl16), apoptosis pathway (Caspase-3, PARP1, CTSD), and TNF signaling pathway (TNFR1, MMP9, ICAM-1, TRAF3) showed significant differential expression starting from 48 h, preceding the NfL increase, suggesting their roles in neuro-axonal damage. Additionally, 11 Alzheimer’s disease-related proteins, with significant changes from 72 to 168 h, were detected only in the proteomic dataset, indicating post-translational modifications might be crucial in neurotoxicity. Pathway analysis revealed that neurodegeneration and Alzheimer’s disease pathways were among the top 15 affected by TMT-induced gene regulation, aligning with the involvement of TNF signaling, apoptosis, and chemokine signaling in neurodegeneration. This research highlighted the value of longitudinal omics studies, combined with pathway enrichment, gene-disease association, and neuro-axonal damage biomarker analyses, to elucidate neurotoxicant-induced neurodegeneration. Findings from this study could enhance the understanding of TMT-induced neurotoxicity, potentially informing future therapeutic strategies and preventive measures.

Treatment with conduritol-β-epoxide (CBE) in preclinical species is expected to be a powerful approach to generate animal models of Gaucher disease (GD) and Parkinson's disease associated with heterozygous mutations in Glucocerebrosidase (GBA-PD). However, it is not fully elucidated how quantitatively the change in glucosylsphingosine (GlcSph) levels in cerebrospinal fluid (CSF) correlates with that in the brain, which is expected to be clinically informative. Herein, we aimed to investigate the correlation with successfully quantified GlcSph in monkey CSF by developing highly sensitive liquid chromatography-tandem mass spectrometry (LC-MS/MS) methods. The GlcSph in normal monkey CSF was 0.635 ± 0.177 pg/mL at baseline and increased by CBE treatment at 3 mg/kg daily for five days up to a moderate level, comparable to that in GD patients. The balance between GlcSph and galactosylsphingosine (GalSph) in the CSF matched that in the brain rather than plasma. In addition, GlcSph in the CSF was increased, accompanied by that in the brain at a dose of 3 mg/kg daily. These results indicate that GlcSph in the CSF is worth evaluating for concentration changes in the brain. Thus, this model can be useful for evaluating GBA-related diseases such as GD and GBA-PD.

Machine learning (ML) approaches have been applied to predicting drug pharmacokinetic properties. Previously, we predicted rat unbound brain-to-plasma ratio (Kpuu,brain) by ML models. In this study, we aimed to predict human Kpuu,brain through animal ML models. First, we re-evaluated ML models for rat Kpuu,brain prediction by using trendy open-source packages. We then developed ML models for monkey Kpuu,brain prediction. Leave-one-out cross validation was utilized to rationally build models using a relatively small dataset. After establishing the monkey and rat ML models, human Kpuu,brain prediction was achieved by implementing the animal models considering appropriate scaling methods. Mechanistic NeuroPK models for the identical monkey and human dataset were treated as the criteria for comparison. Results showed that rat Kpuu,brain predictivity was successfully replicated. The optimal ML model for monkey Kpuu,brain prediction was superior to the NeuroPK model, where accuracy within 2-fold error was 78% (R2 = 0.76). For human Kpuu,brain prediction, rat model using relative expression factor (REF), scaled transporter efflux ratios (ERs), and monkey model using in vitro ERs can provide comparable predictivity to the NeuroPK model, where accuracy within 2-fold error was 71% and 64% (R2 = 0.30 and 0.52), respectively. We demonstrated that ML models can deliver promising Kpuu,brain prediction with several advantages: (1) predict reasonable animal Kpuu,brain; (2) prospectively predict human Kpuu,brain from animal models; and (3) can skip expensive monkey studies for human prediction by using the rat model. As a result, ML models can be a powerful tool for drug Kpuu,brain prediction in the discovery stage.

Multidrug resistance (MDR1) and breast cancer resistance protein (BCRP) play important roles in drug absorption and distribution. Computational prediction of substrates for both transporters can help reduce time in drug discovery. This study aimed to predict the efflux activity of MDR1 and BCRP using multiple machine learning approaches with molecular descriptors and graph convolutional networks (GCNs). In vitro efflux activity was determined using MDR1- and BCRP-expressing cells. Predictive performance was assessed using an in-house dataset with a chronological split and an external dataset. CatBoost and support vector regression showed the best predictive performance for MDR1 and BCRP efflux activities, respectively, of the 25 descriptor-based machine learning methods based on the coefficient of determination (R2). The single-task GCN showed a slightly lower performance than descriptor-based prediction in the in-house dataset. In both approaches, the percentage of compounds predicted within twofold of the observed values in the external dataset was lower than that in the in-house dataset. Multi-task GCN did not show any improvements, whereas multimodal GCN increased the predictive performance of BCRP efflux activity compared with single-task GCN. Furthermore, the ensemble approach of descriptor-based machine learning and GCN achieved the highest predictive performance with R2 values of 0.706 and 0.587 in MDR1 and BCRP, respectively, in time-split test sets. This result suggests that two different approaches to represent molecular structures complement each other in terms of molecular characteristics. Our study demonstrated that predictive models using advanced machine learning approaches are beneficial for identifying potential substrate liability of both MDR1 and BCRP.

Mutations in the GBA gene, encoding glucocerebrosidase (GCase), are linked to Gaucher disease (GD) and are the most common risk factors for Parkinson's disease (PD). The glucosylsphingosine (GlcSph) in cerebrospinal fluid (CSF) is used as a pharmacodynamic marker for GCase functionalizing therapy in GD patients. Its isobaric structural isomer, galactosylsphingosine (GalSph, psychosine), is also used as a diagnostic blood marker in Krabbe disease (KD) which is caused by a deficiency in β-galactocerebrosidase (GALC). However, there are no reports of GlcSph quantification in the CSF of GBA-PD patients and normal healthy humans due to low concentrations. In this study, we successfully quantified GlcSph in healthy human CSF using a highly sensitive LC-MS/MS method with separation of GalSph. The lower limit of quantitation (LLOQ) was 0.1 pg/mL. Additionally, GlcSph and GalSph concentrations in the plasma and brain were determined using different LC-MS/MS methods. The mean concentrations of GlcSph and GalSph in normal human CSF were 1.07 and 9.44 pg/mL, respectively. The GalSph level in the CSF and brain was higher than that of GlcSph, whereas plasma GalSph was lower than GlcSph. Because GCase and GALC are expressed in the brain and the peripheral tissues, GlcSph and GalSph in CSF would be a good surrogate of concentration change in the brain by targeted therapies. This method measures normal levels of GlcSph and GalSph in healthy human CSF without accumulation of sphingolipids, and confirms whether abnormal CSF concentrations can be reduced to normal levels by therapy.

To develop potent and orally bioavailable melatonin receptor (MT1 and MT2) agonists, a novel series of 5-6-5 tricyclic derivatives was designed, synthesized, and evaluated. The synthesized indeno[5,4-d][1,3]oxazole, cyclopenta[c]pyrazolo[1,5-a]pyridine, indeno[5,4-d][1,3]thiazole, and cyclopenta[e]indazole derivatives showed potent binding affinities for MT1/MT2 receptors. Further optimization of these derivatives based on their metabolic stability in human hepatic microsomes revealed that (S)-3b ((S)-N-[2-(2-methyl-7,8-dihydro-6H-indeno[5,4-d][1,3]oxazol-8-yl)ethyl]acetamide) was a potent MT1 and MT2 ligand (MT1, Ki = 0.031 nM; MT2, Ki = 0.070 nM) with good metabolic stability in human hepatic microsomes. Moreover, compound (S)-3b showed good BBB permeability in rats, and its in vivo pharmacological effects were confirmed by its sleep-promotion ability in cats.

The unbound fractions in plasma (f up) in two mouse models of humanized liver mice, PXB and humanized TK-NOG mice, were compared with human f up values using equilibrium dialysis method. A good relationship between f up values obtained from PXB mice and humans was observed; the f up of 34/39 compounds (87.2%) in PXB mice were within 3-fold of human f up. In contrast, a weak correlation was observed between human and humanized TK-NOG mouse f up values; the f up of 15/24 compounds (62.5%) in humanized TK-NOG mice were within 3-fold of human f up. As different profiles of plasma protein binding (PPB) profiles were observed between PXB and humanized TK-NOG mice, f up evaluation is necessary in each mouse model to utilize these humanized liver mice for pharmacological, drug-drug interaction (DDI), and toxicity studies. The unbound fraction in the mixed plasma of human and SCID mouse plasma (85:15) was well correlated with f up in PXB mice (38/39 compounds within a 3-fold). Thus, this artificial PXB mouse plasma could be used to evaluate PPB.

Accurate prediction of oral pharmacokinetics remains challenging. This study investigated quantitative approaches for the prediction of the area under the plasma concentration-time curve after oral administration (AUCp,oral) to rats using the in vitro-in vivo extrapolation (IVIVE), in silico model using machine learning approaches and the combination of the in silico model and in vitro data. A set of 595 structurally diverse compounds with determined AUCp,oral at 1 mg/kg, in vitro intrinsic clearance (CLint), an unbound fraction in plasma (fu,p) in rats, and kinetic solubility at pH 6.8 was used for this assessment. Prediction models developed by two different types of machine learning techniques (i.e., random forest regression and Gaussian processes) were evaluated using three validation methods implementing the time and cluster-split training and test set and fivefold cross-validation. The developed machine learning models have a square of correlation coefficient (R2) in the range of 0.381-0.685 with 33-45% of the compounds being predicted within 2-fold of the observed AUCp,oral value. The predictivity was improved by incorporating CLint, fu,p, and solubility as explanatory variables with R2 = 0.554-0.743. In cases where extraction by the liver is the main elimination pathway and intestinal extraction is negligible, AUCp,oral can be expressed by dose, CLint, and fu,p based on a well-stirred model. By using this conventional IVIVE approach, only 1.7-5.0% of compounds were predicted within the 2-fold error with R2 = 0.354-0.487. Two empirical scaling factors (ESFs) determined by linear regression analysis and machine learning approaches improved the predictivity of AUCp,oral with 33-44% predicted within twofold variability. The IVIVE using ESF predicted by random forest regression showed better predictivity of AUCp,oral with R2 = 0.471-0.618, while it still showed lower predictivity than machine learning approaches applied directly to AUCp,oral prediction. This study demonstrated that the combination of in silico and in vitro parameters is useful to improve the predictivity of the machine learning model for rat AUCp,oral and supports consideration for predicting AUCp,oral for human and other non-clinical species in a similar manner.

The mechanistic neuropharmacokinetic (neuroPK) model was established to predict unbound brain-to-plasma partitioning ( K p,uu,brain ) by considering in vitro efflux activities of multiple drug resistance 1 (MDR1) and breast cancer resistance protein (BCRP). Herein, we directly compare this model to a computational machine learning approach utilizing physicochemical descriptors and efflux ratios of MDR1 and BCRP-expressing cells for predicting K p,uu,brain in rats. Two different types of machine learning techniques, Gaussian processes (GP) and random forest regression (RF), were assessed by the time and cluster-split validation methods using 640 internal compounds. The predictivity of machine learning models based on only molecular descriptors in the time-split dataset performed worse than the cluster-split dataset, whereas the models incorporating MDR1 and BCRP efflux ratios showed similar predictivity between time and cluster-split datasets. The GP incorporating MDR1 and BCRP in the time-split dataset achieved the highest correlation ( R 2 = 0.602). These results suggested that incorporation of MDR1 and BCRP in machine learning is beneficial for robust and accurate prediction. K p,uu,brain prediction utilizing the neuroPK model was significantly worse compared to machine learning approaches for the same dataset. We also investigated the predictivity of K p,uu,brain using an external independent test set of 34 marketed drugs. Compared to machine learning models, the neuroPK model showed better predictive performance with R 2 of 0.577. This work demonstrates that the machine learning model for K p,uu,brain achieves maximum predictive performance within the chemical applicability domain, whereas the neuroPK model is applicable more widely beyond the chemical space covered in the training dataset.

O-GlcNAcase (OGA) has received increasing attention as an attractive therapeutic target for tau-mediated neurodegenerative disorders; however, its role in these pathologies remains unclear. Therefore, potent chemical tools with favorable pharmacokinetic profiles are desirable to characterize this enzyme. Herein, we report the discovery of a potent and novel OGA inhibitor, compound 5i, comprising an aminopyrimidine scaffold, identified by virtual screening based on multiple methodologies combining structure-based and ligand-based approaches, followed by sequential optimization with a focus on ligand lipophilicity efficiency. This compound was observed to increase the level of O-GlcNAcylated protein in cells and display suitable pharmacokinetic properties and brain permeability. Crystallographic analysis revealed that the chemical series bind to OGA via characteristic hydrophobic interactions, which resulted in a high affinity for OGA with moderate lipophilicity. Compound 5i could serve as a useful chemical probe to help establish a proof-of-concept of OGA inhibition as a therapeutic target for the treatment of tauopathies.

Agonistic profiles of AMPA receptor (AMPA-R) potentiators may be associated with seizure risk and bell-shaped dose-response effects. Here, we report the pharmacological characteristics of a novel AMPA-R potentiator, TAK-653, which exhibits minimal agonistic properties. TAK-653 bound to the ligand binding domain of recombinant AMPA-R in a glutamate-dependent manner. TAK-653 strictly potentiated a glutamate-induced Ca 2+ influx in hGluA1i-expressing CHO cells through structural interference at Ser743 in GluA1. In primary neurons, TAK-653 augmented AMPA-induced Ca 2+ influx and AMPA-elicited currents via physiological AMPA-R with little agonistic effects. Interestingly, TAK-653 enhanced electrically evoked AMPA-R-mediated EPSPs more potently than AMPA (agonist) or LY451646 (AMPA-R potentiator with a prominent agonistic effect) in brain slices. Moreover, TAK-653 improved cognition for both working memory and recognition memory, while LY451646 did so only for recognition memory, and AMPA did not improve either. These data suggest that the facilitation of phasic AMPA-R activation by physiologically-released glutamate is the key to enhancing synaptic and cognitive functions, and nonselective activation of resting AMPA-Rs may negatively affect this process. Importantly, TAK-653 had a wide safety margin against convulsion; TAK-653 showed a 419-fold (plasma C max ) and 1017-fold (AUC plasma ) margin in rats. These findings provide insight into a therapeutically important aspect of AMPA-R potentiation.

Capturing unbound drug exposure in the brain is crucial to evaluate pharmacological effects for drugs acting on the central nervous system. However, to date, there are no reports of validated prediction models to determine the brain-to-plasma unbound concentration ratio (Kp,uu,brain) as well as the cerebrospinal fluid (CSF)-to-plasma unbound concentration ratio (Kp,uu,CSF) between humans and other species. Here, we developed a translational CNS steady-state drug disposition model to predict Kp,uu,brain and Kp,uu,CSF across rats, monkeys, and humans by estimating the relative activity factors (RAF) for MDR1 and BCRP in addition to scaling factors (γ and σ) using the molecular weight, logD, CSF bulk flow, and in vitro transport activities of these transporters. In this study, 68, 26, and 28 compounds were tested in the rat, monkey, and human models, respectively. Both the predicted Kp,uu,brain and Kp,uu,CSF values were within the 3-fold range of the observed values (71, 73, and 79%; 79, 88, and 78% of the compounds, respectively), indicating successful prediction of Kp,uu,brain and Kp,uu,CSF in the three species. The overall predictivity of the RAF approach is consistent with that of the relative expression factor (REF) approach. As the established model can predict Kp,uu,brain and Kp,uu,CSF using only in vitro and physicochemical data, this model would help avoid ethical issues related to animal use and improve CNS drug discovery workflow.

The in vitro-in vivo extrapolation (IVIVE) approach for predicting total plasma clearance (CLtot) has been widely used to rank order compounds early in discovery. More recently, a computational machine learning approach utilizing physicochemical descriptors and fingerprints calculated from chemical structure information has emerged, enabling virtual predictions even earlier in discovery. Previously, this approach focused more on in vitro intrinsic clearance (CLint) prediction. Herein, we directly compare these two approaches for predicting CLtot in rats. A structurally diverse set of 1114 compounds with known in vivo CLtot in vitro CLint and plasma protein binding was used as the basis for this evaluation. The machine learning models were assessed by validation approaches using the time- and cluster-split training and test sets, and five-fold cross validation. Assessed by five-fold validation, the random forest regression (RF) and radial basis function (RBF) models demonstrated better prediction performance in eight attempted machine learning models. The CLtot values predicted by the RF and RBF models were within two-fold of the observed values for 67.7 and 71.9% of duster-split test set compounds, respectively, while the predictivity was worse in the time-split dataset. The predictivity of both models tended to be improved by incorporating in vitro parameters, unbound fraction in plasma (f(u,p)), and CLint. CLtot prediction utilizing in vitro CLint and the well-stirred model, correcting for the fraction unbound in blood, was substantially worse compared to machine learning approaches for the same duster-split test set. The reason that CLtot is underestimated by IVIVE is not fully explained by considering the calculated microsomal unbound fraction (cf(u,mic)), extended dearance classification system (ECCS), and omitting high clearance compounds in excess of hepatic blood flow. The analysis suggests that in silico machine learning models may have the power to reduce reliance on or replace in vitro and in vivo studies for chemical structure optimization in early drug discovery.

PURPOSE:Intranasal administration enhances drug delivery to the brain by allowing targeted-drug delivery. Here, we investigated the properties that render a compound suitable for intranasal administration, and the differences between rodents and non-human primates in delivery to the brain.METHODS:The delivery of 10 low-permeable compounds to the brain, including substrates of efflux drug transporters expressed in the blood-brain barrier (didanosine, metformin, zolmitriptan, cimetidine, methotrexate, talinolol, ranitidine, atenolol, furosemide, and sulpiride) and two high-permeable compounds (ropinirole and midazolam) was evaluated following intranasal and intravenous administration in rats. Six of the 12 compounds (metformin, cimetidine, methotrexate, talinolol, sulpiride, and ropinirole) were also evaluated in monkeys, which have a similar nasal cavity anatomical structure to humans.RESULTS:In rats, most of the low-permeable compounds displayed an obvious increase in the brain/plasma concentration ratio (Kp) by intranasal administration (despite their substrate liability for efflux drug transporters); this was not observed with the high-permeable compounds. Similarly, intranasal administration increased Kp for all low-permeable compounds in monkeys.CONCLUSIONS:Compound permeability is a key determinant of Kp increase by intranasal administration. This route of administration is more beneficial for low-permeable compounds and enhances their delivery to the brain in rodents and non-human primates.