Fumiyoshi Yamashita
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Fumiyoshi Yamashita has a total of 138 co-authors. Here are the top co-authors by number of shared publications:
Mitsuru Hashida
Department of Drug Delivery Research, Graduate School of Pharmaceutical Sciences, Kyoto University
15 shared publicationsYuriko Higuchi
Department of Drug Delivery Research, Graduate School of Pharmaceutical Sciences, Kyoto University, Kyoto 606-8501, Japan
9 shared publicationsMasahiro Tsuda
Department of Applied Pharmaceutics and Pharmacokinetics, Graduate School of Pharmaceutical Sciences, Kyoto University, Kyoto 606-8501, Japan
7 shared publications- Mitsuru Hashida
Institute for Advanced Study (KUIAS), Institute for Integrated Cell-Material Sciences (iCeMS), Kyoto University
7 shared publications Kanako So
Department of Quantitative Pharmaceutics, Graduate School of Pharmaceutical Sciences, Kyoto University, Kyoto, Japan
6 shared publications- Shigeru Kawakami
Graduate School of Biomedical Sciences, Nagasaki University, Nagasaki, Japan
4 shared publications
Latest Publications
- 27 October 2026
- Published by Springer Nature in AAPS PharmSciTech
- Vol. 27 (8) , 339
Colorimetric Microplate Assay Coupled with Nonlinear Mixed-Effects Modeling for High-Throughput Evaluation of Ophthalmic Preservatives
- 27 October 2026
- Published by Springer Nature in AAPS PharmSciTech
- Vol. 27 (8) , 339
Colorimetric Microplate Assay Coupled with Nonlinear Mixed-Effects Modeling for High-Throughput Evaluation of Ophthalmic Preservatives
- Open Access
- 12 October 2026
- Published by American Chemical Society (ACS) in Molecular Pharmaceutics
- Vol. 23 (11) , 5385-5398
Prediction of Plasma Concentration−Time Profiles in Mice from Chemical Structure Using a Kinetically Constrained Hybrid Neural ODE
Predicting plasma concentration−time profiles directly from chemical structure remains intrinsically challenging in early drug discovery. Existing approaches typically fall into two paradigms: mechanistic reconstruction via predicted PK parameters, which is constrained by the assumed model structure, and direct data-driven prediction of concentrations, which often suffers from instability and poor temporal coherence under sparse and irregular sampling. Here, we present a kinetically constrained hybrid Neural ordinary differential equation (KCH-NODE) framework for structure-to-PK prediction that bridges these limitations. KCH-NODE anchors predictions to a shared, learnable two-compartment typical trajectory while modeling compound-specific PK behavior as continuous-time deviations governed by a Neural ODE. This hybrid design generates flexible yet pharmacokinetically coherent trajectories, even under sparse observations. Trained on mouse intravenous (IV) and oral (PO) PK data from 1159 compounds across 30 projects, KCH-NODE generally provided more accurate predictions than a purely data-driven Neural ODE in both compound-level and scaffold-based cross-validation (median RMSE 0.374 and 0.388 for IV; 0.564 and 0.570 for PO) and in a temporally held-out evaluation (223 compounds; median RMSE 0.637 for IV and 0.731 for PO), with a smaller loss of accuracy on the temporal split. Predictive accuracy for oral administration nevertheless remained limited on the temporal test set for both methods. Integrated Gradients supported mechanistically consistent feature contributions, and structural similarity to the model-development dataset together with ensemble disagreement provided complementary indicators of prediction reliability. These results indicate that kinetically constrained hybrid Neural ODEs offer an effective and interpretable framework for structure-to-PK profile prediction.
- 12 October 2026
- Published by American Chemical Society (ACS) in Molecular Pharmaceutics
- Vol. 23 (11) , 5385-5398
Open AccessPrediction of Plasma Concentration−Time Profiles in Mice from Chemical Structure Using a Kinetically Constrained Hybrid Neural ODE
Predicting plasma concentration−time profiles directly from chemical structure remains intrinsically challenging in early drug discovery. Existing approaches typically fall into two paradigms: mechanistic reconstruction via predicted PK parameters, which is constrained by the assumed model structure, and direct data-driven prediction of concentrations, which often suffers from instability and poor temporal coherence under sparse and irregular sampling. Here, we present a kinetically constrained hybrid Neural ordinary differential equation (KCH-NODE) framework for structure-to-PK prediction that bridges these limitations. KCH-NODE anchors predictions to a shared, learnable two-compartment typical trajectory while modeling compound-specific PK behavior as continuous-time deviations governed by a Neural ODE. This hybrid design generates flexible yet pharmacokinetically coherent trajectories, even under sparse observations. Trained on mouse intravenous (IV) and oral (PO) PK data from 1159 compounds across 30 projects, KCH-NODE generally provided more accurate predictions than a purely data-driven Neural ODE in both compound-level and scaffold-based cross-validation (median RMSE 0.374 and 0.388 for IV; 0.564 and 0.570 for PO) and in a temporally held-out evaluation (223 compounds; median RMSE 0.637 for IV and 0.731 for PO), with a smaller loss of accuracy on the temporal split. Predictive accuracy for oral administration nevertheless remained limited on the temporal test set for both methods. Integrated Gradients supported mechanistically consistent feature contributions, and structural similarity to the model-development dataset together with ensemble disagreement provided complementary indicators of prediction reliability. These results indicate that kinetically constrained hybrid Neural ODEs offer an effective and interpretable framework for structure-to-PK profile prediction.
- 17 September 2026
- Published by Springer Nature in Pharmaceutical Research
- Vol. 43 (9) , 2957-2966
Federated Estimation of Population Pharmacokinetic Models with Exact Equivalence to Pooled-Data Analysis
Purpose Population pharmacokinetic (PopPK) analysis relies on nonlinear mixed-effects modeling of pooled data; however, sharing subject-level data across sites is often restricted. This study develops a federated framework that exactly recovers the global objective function corresponding to centralized pooled-data analysis, thereby yielding identical parameter estimates without sharing subject-level data. Methods Each site computes its local contributions to the global objective function and corresponding gradients. A central server aggregates these quantities to reconstruct the global objective function and gradient at each parameter value. Performance was evaluated using simulations of a one-compartment oral model under two multi-site scenarios: heterogeneous sparse sampling and between-site heterogeneity in covariate distribution. Federated estimation was compared with centralized pooled-data estimation, independent site-specific estimation, and weighted combination of site-specific estimates and variances. Parameter uncertainty was assessed using bootstrap resampling. Results Across scenarios, federated estimation yielded parameter estimates and bootstrap confidence intervals that closely matched with those from centralized pooled-data estimation for fixed effects, between-subject variability, and residual error. In contrast, independent site-specific estimation showed deviations under certain conditions, and weighted pooling of site-specific estimates did not reproduce centralized variance estimates. Conclusions The proposed federated framework enables mathematically equivalent PopPK estimation without sharing subject-level data.
- 17 September 2026
- Published by Springer Nature in Pharmaceutical Research
- Vol. 43 (9) , 2957-2966
Federated Estimation of Population Pharmacokinetic Models with Exact Equivalence to Pooled-Data Analysis
Purpose Population pharmacokinetic (PopPK) analysis relies on nonlinear mixed-effects modeling of pooled data; however, sharing subject-level data across sites is often restricted. This study develops a federated framework that exactly recovers the global objective function corresponding to centralized pooled-data analysis, thereby yielding identical parameter estimates without sharing subject-level data. Methods Each site computes its local contributions to the global objective function and corresponding gradients. A central server aggregates these quantities to reconstruct the global objective function and gradient at each parameter value. Performance was evaluated using simulations of a one-compartment oral model under two multi-site scenarios: heterogeneous sparse sampling and between-site heterogeneity in covariate distribution. Federated estimation was compared with centralized pooled-data estimation, independent site-specific estimation, and weighted combination of site-specific estimates and variances. Parameter uncertainty was assessed using bootstrap resampling. Results Across scenarios, federated estimation yielded parameter estimates and bootstrap confidence intervals that closely matched with those from centralized pooled-data estimation for fixed effects, between-subject variability, and residual error. In contrast, independent site-specific estimation showed deviations under certain conditions, and weighted pooling of site-specific estimates did not reproduce centralized variance estimates. Conclusions The proposed federated framework enables mathematically equivalent PopPK estimation without sharing subject-level data.
- 9 July 2026
- Published by Springer Nature in Pharmaceutical Research
- Vol. 43 (8) , 1-10
Variance-Consistent Covariate Modeling from Posterior Summaries in Population Pharmacokinetics
Purpose Systematic covariate modeling in nonlinear mixed-effects (NLME) analysis is computationally intensive due to repeated refitting to concentration–time data. Although empirical Bayes estimates (EBEs) facilitate screening, η-shrinkage attenuates between-subject variability and distorts covariance structures, leading to shrinkage bias. We propose a variance-consistent framework enabling covariate modeling from a single base-model fit. Methods The proposed approach incorporates subject-specific posterior means and covariances from an NLME base model. A variance-matching penalty enforces consistency between the total between-subject covariance (model-explained and unexplained) and the base model estimates, preserving the covariance structure without refitting. Performance was compared with EBE regression, two-stage Bayesian estimation, and NLME covariate modeling. Stepwise covariate selection was evaluated using likelihood ratio tests, with the resulting structure compared against the NLME-identified structure as the gold-standard reference. Results Under substantial η-shrinkage of approximately 30%, EBE regression and two-stage Bayesian estimation attenuated covariate-effect parameter estimates. The proposed method provided unbiased estimates, mitigating shrinkage bias and recovering covariate-effect parameter estimates obtained with NLME. It also reproduced NLME-based stepwise covariate selection with high computational scalability by avoiding repeated refitting to time-course data. Conclusions Variance-consistent posterior-based covariate modeling provides a statistically coherent and computationally scalable framework for systematic covariate identification in population PKPD analysis.
- 9 July 2026
- Published by Springer Nature in Pharmaceutical Research
- Vol. 43 (8) , 1-10
Variance-Consistent Covariate Modeling from Posterior Summaries in Population Pharmacokinetics
Purpose Systematic covariate modeling in nonlinear mixed-effects (NLME) analysis is computationally intensive due to repeated refitting to concentration–time data. Although empirical Bayes estimates (EBEs) facilitate screening, η-shrinkage attenuates between-subject variability and distorts covariance structures, leading to shrinkage bias. We propose a variance-consistent framework enabling covariate modeling from a single base-model fit. Methods The proposed approach incorporates subject-specific posterior means and covariances from an NLME base model. A variance-matching penalty enforces consistency between the total between-subject covariance (model-explained and unexplained) and the base model estimates, preserving the covariance structure without refitting. Performance was compared with EBE regression, two-stage Bayesian estimation, and NLME covariate modeling. Stepwise covariate selection was evaluated using likelihood ratio tests, with the resulting structure compared against the NLME-identified structure as the gold-standard reference. Results Under substantial η-shrinkage of approximately 30%, EBE regression and two-stage Bayesian estimation attenuated covariate-effect parameter estimates. The proposed method provided unbiased estimates, mitigating shrinkage bias and recovering covariate-effect parameter estimates obtained with NLME. It also reproduced NLME-based stepwise covariate selection with high computational scalability by avoiding repeated refitting to time-course data. Conclusions Variance-consistent posterior-based covariate modeling provides a statistically coherent and computationally scalable framework for systematic covariate identification in population PKPD analysis.
- 24 August 2025
- Published by Elsevier in International Journal of Pharmaceutics
- Vol. 684, 126111
Harnessing large language models for structured extraction of cytochrome P450–substance interactions from biomedical texts
- 24 August 2025
- Published by Elsevier in International Journal of Pharmaceutics
- Vol. 684, 126111
Harnessing large language models for structured extraction of cytochrome P450–substance interactions from biomedical texts
- Open Access
Harnessing Large Language Models for Structured Extraction of CYP–Substance Interactions from Biomedical Texts
Building on our previous work in biomedical text mining, we revisit the extraction of cytochrome P450 (CYP) and substance interactions using recent advances in large language models (LLMs). We present a scalable, high-accuracy framework that leverages the ChatGPT O3-mini model, employing optimized prompting and batch processing without relying on dictionaries or domain-specific ontologies. Our system achieves strong performance, with recall and precision of 0.950 and 0.978 across all CYP targets, and 0.923 and 0.993 for CYP3A4 specifically. This represents a substantial improvement over our earlier rule-based method. The resulting large-scale analysis not only reflects existing knowledge but also enables a more systematic and comprehensive integration of CYP isoform–substance interaction data, addressing the limitations of previous fragmented efforts. While previous studies have attempted to catalog these interactions, the scale, precision, and automation demonstrated here represent a significant step forward. These findings underscore the potential of LLM-driven pipelines to accelerate biomedical text mining and to support research in drug metabolism and related fields.
Open AccessHarnessing Large Language Models for Structured Extraction of CYP–Substance Interactions from Biomedical Texts
Building on our previous work in biomedical text mining, we revisit the extraction of cytochrome P450 (CYP) and substance interactions using recent advances in large language models (LLMs). We present a scalable, high-accuracy framework that leverages the ChatGPT O3-mini model, employing optimized prompting and batch processing without relying on dictionaries or domain-specific ontologies. Our system achieves strong performance, with recall and precision of 0.950 and 0.978 across all CYP targets, and 0.923 and 0.993 for CYP3A4 specifically. This represents a substantial improvement over our earlier rule-based method. The resulting large-scale analysis not only reflects existing knowledge but also enables a more systematic and comprehensive integration of CYP isoform–substance interaction data, addressing the limitations of previous fragmented efforts. While previous studies have attempted to catalog these interactions, the scale, precision, and automation demonstrated here represent a significant step forward. These findings underscore the potential of LLM-driven pipelines to accelerate biomedical text mining and to support research in drug metabolism and related fields.
- Open Access
- 2 May 2025
- Published by Springer Nature in AAPS PharmSciTech
- Vol. 26 (5) , 1-11
Enhancing Intracellular Uptake of Ivermectin through Liposomal Encapsulation
Ivermectin (IVM), an antiparasitic drug approved by the Food and Drug Administration (FDA), is widely used to treat several neglected tropical diseases, including onchocerciasis, helminthiases, and scabies. Additionally, IVM has shown potential as a potent inhibitor of certain RNA viruses, such as SARS-CoV-2. However, IVM is highly hydrophobic, essentially insoluble in water, which limits its bioavailability and therapeutic effectiveness. The use of liposomes as drug carriers offers several advantages, including enhanced solubility for lipophilic drugs, passive targeting of immune system cells, sustained release, and improved tissue penetration. To address the limitations of IVM, including its poor solubility and bioavailability, liposomal formulations were developed using a combination of soyphosphatidylcholine (SPC), dioleylphosphatidylcholine (DOPC), cholesterol (Ch), and diethylphosphate (DCP) in two distinct molar ratios (1.85:1:0.15 and 7:2:1) via the ethanol injection method. The physicochemical properties of the placebo and IVM-loaded liposomes were extensively characterized in our earlier study, including the particle size, polydispersity index, and zeta potential. The present work adds a deeper level of investigation into how to effect cellular uptake and cytotoxicity in vitro of both free IVM and IVM-loaded liposomes in Vero E6 cells. The half-maximal cytotoxic concentrations (CC50) for free IVM and IVM-loaded liposomes were 10 μM and > 110 μM, respectively and the cellular uptake of IVM-loaded liposomes ranged from 13 to 60%, whereas free IVM showed a significantly lower uptake of only 2%. These results demonstrate that liposomal encapsulation effectively enhances IVM’s cellular uptake while reducing its cytotoxicity, thus offering a promising strategy for improving the effectiveness of IVM. Graphical Abstract
- 2 May 2025
- Published by Springer Nature in AAPS PharmSciTech
- Vol. 26 (5) , 1-11
Open AccessEnhancing Intracellular Uptake of Ivermectin through Liposomal Encapsulation
Ivermectin (IVM), an antiparasitic drug approved by the Food and Drug Administration (FDA), is widely used to treat several neglected tropical diseases, including onchocerciasis, helminthiases, and scabies. Additionally, IVM has shown potential as a potent inhibitor of certain RNA viruses, such as SARS-CoV-2. However, IVM is highly hydrophobic, essentially insoluble in water, which limits its bioavailability and therapeutic effectiveness. The use of liposomes as drug carriers offers several advantages, including enhanced solubility for lipophilic drugs, passive targeting of immune system cells, sustained release, and improved tissue penetration. To address the limitations of IVM, including its poor solubility and bioavailability, liposomal formulations were developed using a combination of soyphosphatidylcholine (SPC), dioleylphosphatidylcholine (DOPC), cholesterol (Ch), and diethylphosphate (DCP) in two distinct molar ratios (1.85:1:0.15 and 7:2:1) via the ethanol injection method. The physicochemical properties of the placebo and IVM-loaded liposomes were extensively characterized in our earlier study, including the particle size, polydispersity index, and zeta potential. The present work adds a deeper level of investigation into how to effect cellular uptake and cytotoxicity in vitro of both free IVM and IVM-loaded liposomes in Vero E6 cells. The half-maximal cytotoxic concentrations (CC50) for free IVM and IVM-loaded liposomes were 10 μM and > 110 μM, respectively and the cellular uptake of IVM-loaded liposomes ranged from 13 to 60%, whereas free IVM showed a significantly lower uptake of only 2%. These results demonstrate that liposomal encapsulation effectively enhances IVM’s cellular uptake while reducing its cytotoxicity, thus offering a promising strategy for improving the effectiveness of IVM. Graphical Abstract
- 16 April 2025
- Published by Elsevier in International Journal of Pharmaceutics
- Vol. 676, 125608
Prediction of plasma concentration-time profiles in mice using deep neural network integrated with pharmacokinetic models
- 16 April 2025
- Published by Elsevier in International Journal of Pharmaceutics
- Vol. 676, 125608
Prediction of plasma concentration-time profiles in mice using deep neural network integrated with pharmacokinetic models
- 19 March 2025
- Published by Elsevier in Toxicology Letters
- Vol. 407, 24-31
Protective role of hepatic non-parenchymal cells against drug-induced hepatocyte toxicity using perfluoropolyether-based microfluidic devices
- 19 March 2025
- Published by Elsevier in Toxicology Letters
- Vol. 407, 24-31
Protective role of hepatic non-parenchymal cells against drug-induced hepatocyte toxicity using perfluoropolyether-based microfluidic devices