Murali Ramanathan

Background: Dysregulated lipid metabolism, which can affect remyelination and neural repair, has not been extensively investigated in progressive multiple sclerosis (PMS). The goal was to examine whether dysregulated cholesterol metabolism was associated with disability in PMS. Methods: The study sample comprised 53 Severe PMS patients (Expanded Disability Status Scale, EDSS score: 7.0–9.5) requiring skilled nursing and 53 age, sex‐ and disease duration‐matched community‐dwelling Control PMS patients (EDSS score: 3.0–6.5). Patients underwent neurological examinations and provided blood samples. Cholesterol and apolipoprotein (Apo) profiles, nuclear magnetic resonance spectroscopy metabolomic profiles, inflammatory cytokines, neuroaxonal, and glial injury biomarkers were obtained. Disability was measured on the EDSS and the Combinatorial Weight‐Adjusted Disability (CombiWISE) scales. Results: The Severe PMS group had lower high‐density lipoprotein cholesterol (HDL‐C), ApoA‐I, and ApoA‐II levels (all p < 0.001) and higher ApoB and ApoE levels. The nutritional status biomarker, albumin, was lower, and the chronic inflammation marker, GlycA, was higher in Severe PMS versus Control PMS (all p < 0.001). Total cholesterol, ApoA‐II, and ApoB were associated with disability on the EDSS and CombiWISE scale. Difference‐correlogram analyses revealed substantial remodeling of biomarker interdependencies in Severe PMS. ApoE showed stronger associations with lipid and nutritional biomarkers in Severe PMS, but with neuroaxonal markers (serum neurofilament light chain, glial fibrillary acidic protein) in Control PMS. Inflammatory cytokines (tumor necrosis factor, interleukin‐6) showed stronger associations with apolipoproteins in Severe PMS, whereas several lipid‐inflammation and lipid nutritional correlations were attenuated. Conclusions: Disrupted cholesterol metabolism and chronic inflammation are strongly linked to greater disability in Severe PMS. Our results also indicate marked alterations in metabolic, inflammatory, nutritional, and neuroglial biomarker networks in Severe PMS.

Purpose: Clinical trial protocols contain critical clinical pharmacology information that is often scattered across different sections of these lengthy documents. The purpose was to evaluate whether a retrieval-augmented generation (RAG) protocol review integrated with knowledge graphs (KGs) could be used to obtain clinical pharmacology evidence and identify regulatory gaps. Methods: Clinical trial protocols were obtained from the Gates Foundation’s publicly available corpus of clinical trials in low- and middle-income countries. The RAG protocol review was used to assess the protocols against Food & Drug Administration (FDA) guidance documents for hepatic impairment, exposure-response, population PK (PopPK), and physiologically based pharmacokinetic modeling (PBPK). Disease-specific KG analyses were conducted for tuberculosis (TB), human immunodeficiency virus (HIV), malaria, and neglected tropical diseases (NTD). KG models connecting studies to drugs, diseases, pharmacologic classes, organ-function exclusions, biomarkers, and pharmacokinetic (PK) parameter features were obtained. Results: Across 42 unique trials, the KG contained 223 nodes and 376 edges. TB protocols produced the largest disease-specific graph, whereas HIV had the highest frequency of PK studies. Gap analysis showed that hepatic impairment exclusions in TB were more frequent than PK studies. In the multi-guidance evaluation, faithfulness was high, answer relevancy was lower, and the clinical pharmacology metric showed guidance-dependent differences. Conclusion: The integrated KG-RAG approach can help highlight potential gaps in clinical pharmacology reporting in infectious disease clinical trial protocols.

To evaluate diffusion model-based artificial intelligence approaches for generating virtual populations with physiological determinants of drug dosing (PDODD) and pharmacokinetic (PK) profiles. A denoising diffusion probabilistic model (DDPM) was applied to a 31-variable dataset of PDODD covariates (18 continuous, 13 binary) from the National Health and Nutrition Examination Survey and compared to a tabular variational autoencoder (TVAE). For nivolumab PK data (12,000 patients, 13 time points, 5 covariates), sequence-based diffusion model (SDM) and a time-aware diffusion model (TDM) with temporal self-attention were evaluated. The predictive performance of the TDM was evaluated by imputing masked time points. All models were trained and tested on 80%:20% partitions of the data using univariate, bivariate, and multivariate distributional similarity metrics. The diffusion model satisfactorily approximated the univariate distributions of continuous PDODD biomarkers (mean Kolmogorov-Smirnov D-statistic, KSD = 0.014), disease status frequencies (mean absolute error, MAE = 0.31%), and preserved bivariate correlations (MAE = 0.033). DDPM outperformed TVAE for categorical variables (0.31% vs. 1.07% MAE) and correlation (0.033 vs. 0.091 MAE). For nivolumab PK, SDM has KSD of 0.047 and a relative error of 1.36%. TDM accurately imputed missing PK timepoints (KSD = 0.014), reconstructing masked Day 1, Peak concentration (Cmax) Dose-9, and Terminal phase concentrations with MAE of 0.43%, 0.25%, and 0.76%, and correlations ≥ 0.999. Diffusion models demonstrated strong performance in generating cross-sectional PK covariate data and longitudinal PK profiles, capturing complex distributional and temporal dependencies. Diffusion-based approaches provide a flexible and robust framework for virtual simulations in pharmacometrics.

The purpose was to evaluate retrieval‐augmented generative (RAG) artificial intelligence (AI) methods for assessing the regulatory compliance of drug information and adherence to best practices in clinical trial protocols. Integrated systems containing RAG and large language model (LLM) components were employed to evaluate drug information and clinical trial protocols. The drug information for adalimumab, insulin glargine, atorvastatin calcium, sertraline, and alprazolam was evaluated for compliance with Food and Drug Administration (FDA) clinical pharmacology guidance for indications, use in specific populations, and warnings and precautions. The reasons for the withdrawal of rofecoxib, valdecoxib, and troglitazone were elicited. The clinical trial protocol evaluation system was used to assess a Phase‐2a clinical trial protocol of Rifafour in tuberculosis with the FDA E9 and E9 (R1) guidance documents. The RAG system correctly identified the indications, use in specific populations, and warnings and precautions for adalimumab, insulin glargine, atorvastatin calcium, sertraline, and alprazolam. The drug information was evaluated against the requirements in the guidance documents, confirming compliance when present and providing explanations for deficiencies. The causes underlying the withdrawal of rofecoxib, valdecoxib, and troglitazone were explained. The clinical protocol summary included study design, population definitions, treatments, dose levels, and route of administration. The summary of the statistical analysis plan included primary/secondary endpoints, statistical tests, pharmacokinetic parameters, and handling of missing data and outliers. The findings aligned with manual protocol reviews. RAG‐based AI methods can improve the usefulness of LLMs in document‐restricted settings and are a promising approach for evaluating the compliance of clinical pharmacology documents.

To evaluate the strengths and limitations of retrieval-augmented generative (RAG) artificial intelligence (AI) for natural language querying of biologics immunogenicity data. The package inserts for drugs approved with biologics license applications (BLA) were retrieved from DailyMed ( https://dailymed.nlm.nih.gov/dailymed/ ). The RAG system integrated natural language processing, retrieval, and large language model (LLM) components. ChatGPT, Gemini, DeepSeek, and Llama were queried with five clinical pharmacology-focused questions on factors influencing anti-drug antibody (ADA) incidence and tolerability, including effects of target protein, administration route, and citrate excipients. Outputs were assessed for relevance, faithfulness, and domain-specific accuracy. The dataset included 663 biologics, of which 206 (31.1%) were monoclonal antibodies. The RAG system retrieved relevant contexts for all queries, but several contexts contained inaccuracies related to the presence of non-antibody protein drugs. All four LLMs generated coherent summaries and identified determinants of ADA incidence, such as drug type, assay methods, and concomitant therapy. All models found that injection-site pain occurred with some protein therapeutics containing citrate excipients, and that evidence for a direct causal role of citrate was mixed. Comparative evaluation showed that LLM outputs were generally relevant and faithful to the source text, with variation in the level of detail and comprehensiveness across models. Domain-specific evaluations indicated that responses accurately identified trends in immunogenicity and highlighted the knowledge gaps. While RAG-based systems can retrieve and synthesize immunogenicity assessments from multiple source documents, significant limitations were noted in this use case. The effectiveness of the retriever can limit RAG performance and warrant refinement. Graphical Abstract

Purpose: To investigate associations between lipid oxidation biomarkers (oxylipins), antioxidant micronutrients, lipoprotein particles, and apolipoproteins in multiple sclerosis (MS). Methods: Blood and neurological assessments were collected from 30 healthy controls, 68 relapsing remitting MS subjects, and 37 progressive MS subjects. Hydroxy (H) and hydroperoxy lipid peroxidation products of the polyunsaturated fatty acids (PUFAs) arachidonic (20:4, ω-6), linoleic (octadecadienoic acid or ODE, 18:2, ω-6), eicosapentaenoic (20:5, ω-3), and α-linolenic (18:3, ω-3) acids were measured using liquid chromatography–mass spectrometry. Antioxidant micronutrients, including β-cryptoxanthin and lutein/zeaxanthin, were quantified by high-performance liquid chromatography. Lipoprotein and metabolite profiles were obtained using nuclear magnetic resonance spectroscopy. Regression models were adjusted for age, sex, body mass index, and disease status. Results: The 9-hydroxy octadecadienoic acid to 13-hydroxy octadecadienoic acid ratio (9-HODE/13-HODE ratio), which reflects autoxidative versus enzymatic oxidation, was associated with MS status (p = 0.002) and disability on the Expanded Disability Status Scale (p = 0.004). Lutein/zeaxanthin (p = 0.023) and β-cryptoxanthin (p = 0.028) were negatively associated with the 9-HODE/13-HODE ratio. Apolipoprotein-CII, a marker of liver-X-receptor (LXR) signaling, was associated with 9-HODE/13-HODE ratio and other oxylipins. Octadecadienoic fatty acid-derived oxylipins were negatively associated with LC3A, a mitophagy marker, and positively correlated with 7-ketocholesterol, a cholesterol autoxidation product. Conclusions: Autoxidation of PUFAs is associated with greater disability in MS. Higher β-cryptoxanthin and lutein/zeaxanthin were associated with reduced auto-oxidation. Lipid peroxidation shows associations with LXR signaling, mitophagy, inflammation, and cholesterol autoxidation.

Background: To evaluate frailty in severe progressive multiple sclerosis (PMS) and to investigate the underlying mechanisms. Methods: This prospective, cross‐sectional, multicenter study enrolled a late severe PMS group requiring skilled nursing ( n = 53) and an age, sex, and disease duration‐matched control PMS group ( n = 53). Participants received neurological and MRI assessments and provided blood samples. Frailty was measured on the Edmonton Frail Scale. Disability was measured on the Expanded Disability Status Scale (EDSS), and fatigue was assessed on the Fatigue Severity Scale. The inflammatory vulnerability index (IVX) and metabolic vulnerability index (MVX) were computed from nuclear magnetic resonance spectroscopy‐derived metabolomic profiling. Serum neurofilament (sNfL), glial fibrillary acidic protein (GFAP), and growth differentiation factor 15 (GDF15) levels were obtained. Results: The late severe PMS group had a higher median EDSS (8.0 vs. 6.0, p < 0.001) than the matched control PMS group. The late severe PMS group had a higher prevalence of frailty (73.1% vs. 23.1%, p < 0.001) and higher frailty scores (8.87 vs. 5.52, p < 0.001) than the control PMS group. EFS was associated with EDSS in both PMS groups. Positive frailty status was associated with a 1.19‐point greater EDSS ( p = 0.012) in the control PMS group and a 0.436‐point greater EDSS in the late severe PMS group ( p = 0.002). In PMS controls, the EFS and frailty status were associated with IVX ( p = 0.044 for EFS) and MVX ( p = 0.036 for EFS). Conclusions: Frailty is positively associated with MS disability. Inflammatory and metabolic vulnerability are associated with frailty in PMS.

Metabolic dysfunction‐associated fatty liver disease causes hepatic fat accumulation (steatosis) and fibrosis, which can be measured with ultrasound elastography imaging. The dependencies of elastography‐derived hepatic steatosis and fibrosis measures with chronic inflammation, disease states, and physiological determinants of drug dosing (PDODD) were assessed. Liver elastography data for n = 5494 participants (50% female, 12–80 years) were obtained from the National Health and Nutrition Examination Survey. Controlled attenuation parameter (CAP) and median liver stiffness (LSM) elastography metrics were used to assess steatosis and fibrosis, and their associations with over 50 key organ systems, disease, and PDODD biomarkers were evaluated with statistical regression, ensemble, and Bayesian learning methods. CAP and LSM increased with age and were greater in males, active liver disease, active hepatitis C, and diabetes or prediabetes. LSM was greater in the presence of congestive heart failure and dialysis. The inflammatory markers C‐reactive protein (CRP) and ferritin, body surface area, and hepatic R ‐value were greater in steatosis and fibrosis. Plasma volume, neutrophil, red blood cell, and platelet counts were greater in steatosis. Drug‐induced liver injury index was lower in steatosis and greater in fibrosis. Albumin levels and platelet counts were lower, but the urine albumin‐to‐creatine ratio was greater in fibrosis. Ensemble learning identified interactions among BMI, age, CRP, ferritin, and liver enzymes contributing to steatosis and fibrosis. Bayesian networks were used to identify directed acyclic graph structures for steatosis and fibrosis. Elastography‐derived measures may be useful for individualizing dosing regimens in the presence of metabolic comorbidities presenting dose‐selection challenges.

The purpose was to evaluate the utility of large language models (LLMs) for reviewing the statistical analysis plan (SAP) and pharmacokinetics–pharmacodynamics (PK–PD) components of clinical trial protocols. Clinical trial protocols and SAPs were obtained from clinicaltrials.gov for a testbed of 15 small‐molecule drugs, biologics, and global antibiotic and public health interventions. The GPT‐4o (ChatGPT) LLM was used to elicit study design attributes, relevant guidelines, and detailed SAP evaluations with prompts engineered to the persona of a regulatory expert. The SAP methodology was assessed against the Food and Drug Administration’s (FDA) E9 Statistical Principles for Clinical Trials guidance. The SAP evaluation outputs were assessed in post hoc analyses with ChatGPT and Grok, based on a rubric that evaluated the accuracy of primary outcome identification, the correctness of statistical methodology, compliance with the FDA E9 guidance, and clinical interpretability. PK–PD analysis plans were assessed on the accuracy of PK–PD objectives and measures and PK analysis methods. ChatGPT accurately identified the disease, intervention, and comparator groups for all trials, as well as the study sample size for 14 out of 15 trials. The most frequently cited guidelines were the FDA’s E9 guidance for SAP and the FDA Guidance for Industry: Population Pharmacokinetics for PK–PD. ChatGPT outputs of the SAP and PK–PD analysis plans were clear and organized, demonstrating a satisfactory ability to extract and summarize technical details; some limitations in contextual accuracy were observed. LLMs can be effective tools for assessing the SAP, PK–PD, and other aspects of clinical trial protocol reviews.

Model-informed drug development (MIDD) is a framework that utilizes quantitative modeling and simulation methods to integrate nonclinical and clinical data, as well as prior knowledge, to inform drug development and regulatory decision-making. This scoping review examines the current MIDD landscape with a focus on pharmacometrics. It will clarify essential MIDD concepts and provide a taxonomy of terms. Additionally, it will evaluate the implications of the International Council for Harmonisation (ICH) M15 MIDD draft guidelines and summarize case studies that illustrate the use of MIDD in various drug development and lifecycle management scenarios. The diverse case studies span various therapeutic areas, small molecules, monoclonal antibodies, and long-acting injectable dosage forms. The ICH M15 MIDD guidelines aim to align the expectations of regulators and sponsors, support consistent regulatory decisions, and minimize errors in the acceptance of modeling and simulation. MIDD is a transformative and rigorous process that fosters collaboration between industry and regulatory agencies during drug development.MIDD has enabled accelerated approvals of drugs for pediatric conditions and rare diseases, where it can be difficult to recruit enough patients for efficacy studies. MIDD has contributed to the clinical pharmacology strategies that have successfully allowed dose extrapolation to related disease indications, dosage forms, and clinical populations without additional clinical trials.