Absorption Distribution Metabolism Excretion and Toxicity Property Prediction Utilizing a Pre-Trained Natural Language Processing Model and Its Applications in Early-Stage Drug Development.

Machine learning techniques are extensively employed in drug discovery, with a significant focus on developing QSAR models that interpret the structural information of potential drugs. In this study, the pre-trained natural language processing (NLP) model, ChemBERTa, was utilized in the drug discovery process. We proposed and evaluated four core model architectures as follows: deep neural network (DNN), encoder, concatenation (concat), and pipe. The DNN model processes physicochemical properties as input, while the encoder model leverages the simplified molecular input line entry system (SMILES) along with NLP techniques. The latter two models, concat and pipe, incorporate both SMILES and physicochemical properties, operating in parallel and with sequential manners, respectively. We collected 5238 entries from DrugBank, including their physicochemical properties and absorption, distribution, metabolism, excretion, and toxicity (ADMET) features. The models' performance was assessed by the area under the receiver operating characteristic curve (AUROC), with the DNN, encoder, concat, and pipe models achieved 62.4%, 76.0%, 74.9%, and 68.2%, respectively. In a separate test with 84 experimental microsomal stability datasets, the AUROC scores for external data were 78% for DNN, 44% for the encoder, and 50% for concat, indicating that the DNN model had superior predictive capabilities for new data. This suggests that models based on structural information may require further optimization or alternative tokenization strategies. The application of natural language processing techniques to pharmaceutical challenges has demonstrated promising results, highlighting the need for more extensive data to enhance model generalization.

Application of physiologically-based pharmacokinetic model approach to predict pharmacokinetics and drug-drug interaction of rivaroxaban: A case study of rivaroxaban and carbamazepine.

Rivaroxaban (RIV; Xarelto; Janssen Pharmaceuticals, Beerse, Belgium) is one of the direct oral anticoagulants. The drug is a strong substrate of cytochrome P450 (CYP) enzymes and efflux transporters. This study aimed to develop a physiologically-based pharmacokinetic (PBPK) model for RIV. It contained three hepatic metabolizing enzyme reactions (CYP3A4, CYP2J2, and CYP-independent) and two active transporter-mediated transfers (P-gp and BCRP transporters). To illustrate the performance of the developed RIV PBPK model on the prediction of drug-drug interactions (DDIs), carbamazepine (CBZ) was selected as a case study due to the high DDI potential. Our study results showed that CBZ significantly reduces the exposure of RIV. The area under the concentration-time curve from zero to infinity (AUCinf ) of RIV was reduced by 35.2% (from 2221.3 to 1438.7 ng*h/ml) and by 25.5% (from 2467.3 to 1838.4 ng*h/ml) after the first dose and at the steady-state, respectively, whereas the maximum plasma concentration (Cmax ) of RIV was reduced by 37.7% (from 266.3 to 166.1 ng/ml) and 36.4% (from 282.3 to 179.5 ng/ml), respectively. The developed PBPK model of RIV could be paired with PBPK models of other interested perpetrators to predict DDI profiles. Further studies investigating the extent of DDI between CBZ and RIV should be conducted in humans to gain a full understanding of their safety and effects.

Fine-tuning of BERT Model to Accurately Predict Drug-Target Interactions.

The identification of optimal drug candidates is very important in drug discovery. Researchers in biology and computational sciences have sought to use machine learning (ML) to efficiently predict drug-target interactions (DTIs). In recent years, according to the emerging usefulness of pretrained models in natural language process (NLPs), pretrained models are being developed for chemical compounds and target proteins. This study sought to improve DTI predictive models using a Bidirectional Encoder Representations from the Transformers (BERT)-pretrained model, ChemBERTa, for chemical compounds. Pretraining features the use of a simplified molecular-input line-entry system (SMILES). We also employ the pretrained ProBERT for target proteins (pretraining employed the amino acid sequences). The BIOSNAP, DAVIS, and BindingDB databases (DBs) were used (alone or together) for learning. The final model, taught by both ChemBERTa and ProtBert and the integrated DBs, afforded the best DTI predictive performance to date based on the receiver operating characteristic area under the curve (AUC) and precision-recall-AUC values compared with previous models. The performance of the final model was verified using a specific case study on 13 pairs of subtrates and the metabolic enzyme cytochrome P450 (CYP). The final model afforded excellent DTI prediction. As the real-world interactions between drugs and target proteins are expected to exhibit specific patterns, pretraining with ChemBERTa and ProtBert could teach such patterns. Learning the patterns of such interactions would enhance DTI accuracy if learning employs large, well-balanced datasets that cover all relationships between drugs and target proteins.

A Deep Neural Network-Based Model Predicting Peritumoral Edema After Radiosurgery for Meningioma.

BACKGROUND: Gamma Knife radiosurgery (GKS) is a promising treatment option for meningioma. However, the incidence of peritumoral edema (PTE) following GKS has been reported to be 7%-38%. This study aimed to develop a predictive model for post-GKS PTE using a deep neural network (DNN) algorithm. METHODS: Patients treated with GKS for meningioma between November 2012 and February 2020 at a single tertiary center were reviewed. The primary outcome was newly developed or aggravated PTE after GKS. Clinical data, including radiosurgical parameters, were collected, and imaging data obtained at the time of GKS were incorporated into the model using a 50-layered residual neural network, ResNet50. Consequently, the model efficiency was evaluated considering the accuracy and area under the receiver operating curve (AUC) values. RESULTS: A total of 202 patients were included in this study. The median tumor volume was 2.3 mL, and the median prescription dose was 13 Gy. PTE was observed before GKS in 22 patients. Post-GKS PTE was evident in 28 patients (13.9%), which further evolved to radiation necrosis in 5 patients. The accuracy and AUC values of the hybrid data model based on both clinical and imaging data were 0.725 and 0.701, respectively. The performance of the hybrid data model was superior to that of the other models based on clinical or image data only. CONCLUSIONS: The DNN-based model using both clinical and imaging data exhibited fair results in predicting post-GKS PTE in meningioma treatment. Predictive models using imaging data may be helpful in prognostic research.

Population Pharmacokinetic Method to Predict Within-Subject Variability Using Single-Period Clinical Data.

Sample sizes for single-period clinical trials, including pharmacokinetic studies, are statistically determined by within-subject variability (WSV). However, it is difficult to determine WSV without replicate-designed clinical trial data, and statisticians typically estimate optimal sample sizes using total variability, not WSV. We have developed an efficient population-based method to predict WSV accurately with single-period clinical trial data and demonstrate method performance with eperisone. We simulated 1000 virtual pharmacokinetic clinical trial datasets based on single-period and dense sampling studies, with various study sizes and levels of WSV and interindividual variabilities (IIVs). The estimated residual variability (RV) resulting from population pharmacokinetic methods were compared with WSV values. In addition, 3 × 3 bioequivalence results of eperisone were used to evaluate method performance with a real clinical dataset. With WSV of 40% or less, regardless of IIV magnitude, RV was well approximated by WSV for sample sizes greater than 18 subjects. RV was underestimated at WSV of 50% or greater, even with datasets having low IIV and numerous subjects. Using the eperisone dataset, RV was 44% to 48%, close to the true value of 50%. In conclusion, the estimated RV accurately predicted WSV in single-period studies, validating this method for sample size estimation in clinical trials.

Clinical Evaluation of Acetaminophen-Galgeuntang Interaction Based on Population Approaches.

Galgeuntang (GGT), a traditional herbal medicine, is widely co-administered with acetaminophen (AAP) for treatment of the common cold, but this combination has not been the subject of investigation. Therefore, we investigated the herb-drug interaction between GGT and AAP by population pharmacokinetics (PKs) modeling and simulation studies. To quantify PK parameters and identify drug interactions, an open label, three-treatment, three-period, one-sequence (AAP alone, GGT alone, and AAP and GGT in combination) clinical trial involving 12 male healthy volunteers was conducted. Ephedrine (EPD), the only GGT component detected, was identified using a one-compartment model. The PKs of AAP were described well by a one-compartment model and exhibited two-phase absorption (rapid followed by slow) and first-order elimination. The model showed that EPD significantly influenced the PKs of AAP. The simulation results showed that at an AAP dose of 1000 mg × 4 times daily, the area under the concentration versus time curve of AAP increased by 16.4% in the presence of GGT compared to AAP only. In conclusion, the PKs of AAP were affected by co-administration of GGT. Therefore, when AAP is combined with GGT, adverse effects related to overdose of AAP could be induced possibly.