C. Anthony Hunt

BACKGROUND Chronic pain is a complex, multidimensional experience inadequately captured by traditional clinical measures. Understanding an individual's lived pain experience (LPE) is crucial for personalized treatment, but current methods for capturing LPEs—such as in-depth qualitative interviews—are resource-intensive and difficult to scale. Large language models (LLMs) offer a promising solution by enabling efficient, reproducible synthesis of personalized LPE narratives (LPE-Stories) from interview data and clinical records. OBJECTIVE This pilot study tests the feasibility and acceptability of an LLM-facilitated protocol to help individuals with chronic low back pain (cLBP) develop, validate, and refine a detailed, personally useful LPE-Story. Secondary objectives include evaluating whether LLM-empowered workflows can surface latent information not captured by traditional measures, and assessing practical benefits for participants, their healthcare providers, and their social support networks. METHODS This single-arm, mixed-methods pilot will recruit up to 30 cLBP participants (in groups of five) who recently completed a large observational study. Following an orientation session, participants complete 3 cycles of semi-structured Zoom interviews, each followed by LLM-assisted workflows. In Cycle 1, a workflow generates "Data-Story" from quantitative and categorical data; the participant critiques the Data-Story and begins sharing their LPE, which LLM workflows use to produce an enriched LPE-Story. Cycles 2 and 3 involve iterative review and refinement. An exit interview assesses accuracy, usefulness, and shareability of the final LPE-Story; feasibility is evaluated via engagement, satisfaction, and a 20-item questionnaire RESULTS We have received Institutional Review Board approval. Recruitment and workflow runs are un-derway; results to follow. CONCLUSIONS This pilot will provide preliminary evidence on the feasibility, acceptability, and clinical utility of LLM-empowered workflows for capturing individualized LPEs. If successful, the approach could offer a scalable, reproducible method for generating rich, participant-validated pain narratives that complement traditional clinical measures—informing larger studies aimed at integrating LPE-Stories into clinical practice, enhancing patient-provider communication, and improving personalized pain management.

Objective: This study compares multiple LLMs, including ChatGPT, DeepSeek, and Llama, to generate meaningful, audience-adapted labels for the existing latent classes among patients with chronic low back pain (cLBP). Methods: Phenotypes were derived from baseline data from two cohorts within the NIH HEAL BACPAC consortium: BACKHOME, a large nationwide e-cohort (train set: N = 3025), and COMEBACK, a deep phenotyping cohort (test set: N = 450). The analysis included pain characteristics, psychosocial factors, lifestyle habits, and social determinants of health. ChatGPT-4o (OpenAI), DeepSeek-R1, and Llama 3.3 (Meta) were applied to generate class labels for each combination of audience (clinician, patient, and caregiver), tone (formal, empathetic, and informal), and technicality (high, medium, and low). Results: Latent Class Model (LCM) identified four distinct behavioral phenotypes in patients with cLBP: High Distress and Maladaptive Behaviors, Resilient and Adaptive Coping, Intermediate Maladaptive Patterns, and Emotionally Regulated with High Pain Burden. Previously validated by domain experts, these profiles served as the basis for automated labeling using three LLMs (ChatGPT-4o, DeepSeek-R1, and Llama 3.3). Using different tones and complexity levels, each model produced class labels specific to clinicians, patients, and caregivers. The generated class names for all LLMs closely matched expert-defined traits like emotional regulation, resilience, and high distress, indicating strong conceptual alignment and the capacity of LLMs to generate precise, audience-specific labels for intricate behavioral and psychological profiles. Conclusions: These results highlight the possibility of integrating LLM-driven labeling into research and clinical practice, helping to achieve more transparent knowledge translation, improved decision-making, and personalized care.

Chronic low back pain (LBP) is influenced by a broad spectrum of patient‐specific factors as codified in domains of the biopsychosocial model (BSM). Operationalizing the BSM into research and clinical care is challenging because most investigators work in silos that concentrate on only one or two BSM domains. Furthermore, the expanding, multidisciplinary nature of BSM research creates practical limitations as to how individual investigators integrate current data into their processes of generating impactful hypotheses. The rapidly advancing field of artificial intelligence (AI) is providing new tools for organizing knowledge, but the practical aspects for how AI may advance LBP research and clinical are beginning to be explored. The goals of the work presented here are to: (1) explore the current capabilities of knowledge integration technologies (large language models (LLM), similarity graphs (SGs), and knowledge graphs (KGs)) to synthesize biomedical literature and depict multimodal relationships reflected in the BSM, and; (2) highlight limitations, implementation details, and future areas of research to improve performance. We demonstrate preliminary evidence that LLMs, like GPT‐3, may be useful in helping scientists analyze and distinguish cLBP publications across multiple BSM domains and determine the degree to which the literature supports or contradicts emergent hypotheses. We show that SG representations and KGs enable exploring LBP's literature in novel ways, possibly providing, trans‐disciplinary perspectives or insights that are currently difficult, if not infeasible to achieve. The SG approach is automated, simple, and inexpensive to execute, and thereby may be useful for early‐phase literature and narrative explorations beyond one's areas of expertise. Likewise, we show that KGs can be constructed using automated pipelines, queried to provide semantic information, and analyzed to explore trans‐domain linkages. The examples presented support the feasibility for LBP‐tailored AI protocols to organize knowledge and support developing and refining trans‐domain hypotheses.

Predictions of xenobiotic hepatic clearance in humans using in vitro-to-in vivo extrapolation methods are frequently inaccurate and problematic. Multiple strategies are being pursued to disentangle responsible mechanisms. The objective of this work is to evaluate the feasibility of using insights gained from independent virtual experiments on two model systems to begin unraveling responsible mechanisms. The virtual culture is a software analog of hepatocytes in vitro, and the virtual human maps to hepatocytes within a liver within an idealized model human. Mobile objects (virtual compounds) map to amounts of xenobiotics. Earlier versions of the two systems achieved quantitative validation targets for intrinsic clearance (virtual culture) and hepatic clearance (virtual human). The major difference between the two systems is the spatial organization of the virtual hepatocytes. For each pair of experiments (virtual culture, virtual human), hepatocytes are configured the same. Probabilistic rules govern virtual compound movements and interactions with other objects. We focus on highly permeable virtual compounds and fix their extracellular unbound fraction at one of seven values (0.05–1.0). Hepatocytes contain objects that can bind and remove compounds, analogous to metabolism. We require that, for a subset of compound properties, per-hepatocyte compound exposure and removal rates during culture experiments directly predict corresponding measures made during virtual human experiments. That requirement serves as a cross-system validation target; we identify compound properties that enable achieving it. We then change compound properties, ceteris paribus, and provide model mechanism-based explanations for when and why measures made during culture experiments under- (or over-) predict corresponding measures made during virtual human experiments. The results show that, from the perspective of compound removal, the organization of hepatocytes within virtual livers is more efficient than within cultures, and the greater the efficiency difference, the larger the underprediction. That relationship is noteworthy because most in vitro-to-in vivo extrapolation methods abstract away the structural organization of hepatocytes within a liver. More work is needed on multiple fronts, including the study of an expanded variety of virtual compound properties. Nevertheless, the results support the feasibility of the approach and plan.

We explore a virtual experiment strategy to discover plausible mechanism-based explanations for frequently inaccurate predictions of hepatic drug clearance made using established in vitro-in vivo extrapolation methods. We describe a three-protocol plan that uses validated software analogs of real in vitro and in vivo systems constructed using a common set of objects, components, spaces, and methods. Both systems utilize identical quasi-autonomous hepatocyte analogs containing enzyme-like objects. We parameterize concrete mobile objects (virtual compounds) to simulate the referent drug’s disposition and removal characteristics in vitro and in vivo. The goal of Protocol one (Protocol two) is that measures of virtual compound removal using the in vitro analog (in vivo analog) map directly to the measurements used to compute intrinsic clearance (in vivo hepatic clearance). Protocol three, the focus of this work, requires achieving an essential cross-system validation target. For a subset of virtual compounds, measurements of unbound compounds entering hepatocytes (and their subsequent removal) during virtual in vitro experiments will directly predict corresponding measures made during virtual in vivo experiments. We study four highly permeable virtual compounds when their unbound fraction is fixed at one of seven values (0.05-1.0). Results span the range of hepatic extraction ratios. We achieve the cross-system validation target in 15 cases. In the other 13 cases, explanations of the in vitro-in vivo differences in disposition and removal trace to differences in compound-hepatocyte access within the two analogs during execution. The hepato-mimetic structural organization of hepatocytes within the in vivo analog, which is absent within the in vitro analog, is the determining factor. The results taken together support the feasibility of using the three-protocol plan to help explain observed in vitro-in vivo extrapolation discrepancies. We conjecture that, for some cases, the model mechanism-based explanations of discrepancies described herein will have wet-lab counterparts.

Interpretations of elevated blood levels of alanine aminotransferase (ALT) for drug-induced liver injury often assume that the biomarker is released passively from dying cells. However, the mechanisms driving that release have not been explored experimentally. The usefulness of ALT and related biomarkers will improve by developing mechanism-based explanations of elevated levels that can be expanded and elaborated incrementally. We provide the means to challenge the ability of closely related model mechanisms to generate patterns of simulated hepatic injury and ALT release that scale (or not) to be quantitatively similar to the wet-lab validation targets, which are elevated plasma ALT values following acetaminophen (APAP) exposure in mice. We build on a published model mechanism that helps explain the generation of characteristic spatiotemporal features of APAP hepatotoxicity within hepatic lobules. Discrete event and agent-oriented software methods are most prominent. We instantiate and leverage a small constellation of concrete model mechanisms. Their details during execution help bring into focus ways in which particular sources of uncertainty become entangled with cause-effect details within and across several levels. We scale ALT amounts in virtual mice directly to target plasma ALT values in individual mice. A virtual experiment comprises a set of Monte Carlo simulations. We challenge the sufficiency of four potentially explanatory theories for ALT release. The first of the tested model theories failed to achieve the initial validation target, but each of the three others succeeded. Results for one of the three model mechanisms matched all target ALT values quantitatively. It explains how ALT externalization is the combined consequence of lobular-location-dependent drug-induced cellular damage and hepatocyte death. Falsification of one (or more) of the model mechanisms provides new knowledge and incrementally shrinks the constellation of model mechanisms. The modularity and biomimicry of our explanatory models enable seamless transition from mice to humans. Interpretations of elevated biomarkers for drug-induced liver injury assume passive release during hepatocyte death, yet indirect evidence indicates that plasma levels can increase absent injury. Limitations on measurements make it infeasible to resolve causal linkages between drug disposition and plasma levels of biomarkers. To improve explanatory knowledge, we instantiate within virtual mice, plausible mechanism-based causal linkages between acetaminophen disposition and alanine aminotransferase (ALT) behavior that enables simulation results to meet stringent quantitative validation prerequisites. We challenge the sufficiency of four model mechanisms by scaling ALT amounts in virtual mice to corresponding plasma values. Virtual experiment results in which ALT externalization is a combined consequence of lobular-location-dependent hepatocyte death and drug-induced cellular damage, matches all validation targets. We assert that the actual mechanisms responsible for plasma ALT values in individual mice and the virtual causal processes occurring during model execution are strongly analogous within and among real hepatic lobular levels.

Studying biological systems is difficult because of complexity, variability, and uncertainty. Conceptual models and diagrams are useful in conveying ideas about how a biological phenomenon are thought to be generated. However, sophisticated modeling and simulation methods are needed to discover mechanism-based explanations. Presented herein is a new and unique methodology for this application. Using virtual experiment methods, we recently provided a plausible solution to a problem that had eluded and perplexed pharmacologists and toxicologists for more than 40 years. We describe how virtual and real-world experimentation can be complementary, and propose a way to partially automate the methodology to expedite research.

The in vitro-in vivo extrapolation (IVIVE) methods used currently to predict the hepatic clearance of new chemical entities are plagued by poorly understood inaccuracies. To begin identifying plausible sources, we challenge two of core hypotheses. Hypothesis-1: the intralobular micro-anatomical organization of hepatocytes (HPCs) can be abstracted away. By accepting that hypothesis, one can assume that intrinsic clearance per HPC is essentially the same in vitro and in vivo, and thus an IVIVE method can employ a simplified liver model, typically the "well-stirred" liver model. Hypothesis-2: when the simplified liver model is the "parallel tube model," drug concentration decreases exponentially from portal to central vein. When either simplified liver model is used, a core assumption is that intrinsic clearance is directly proportional to the unbound fraction of drug. A barrier to progress has been the fact that it is currently infeasible to challenge the two hypotheses using wet-lab experiments. In this work, we challenge virtual counterparts of the two hypotheses by experimenting on virtual mice in which hepatic disposition and clearance are consequences of concretized model mechanisms that have met several demanding requirements, including the following. The virtual liver's structure and organization are strongly analogous to those of an actual liver, and the hepatic disposition and clearance of several virtual compounds have achieved quantitative validation targets. We study two virtual compounds. Compound-1 simulates the extreme of low-clearance, highly permeable compounds. Compound-2 simulates a highly permeable compound exhibiting maximum intrinsic clearance. We simulate changes in unbound fraction by changing the probability (pEnter) that a Compound-1 or-2 will enter an adjacent HPC during a simulation cycle. Compound-1 and-2 HPC

Interpretations of elevated blood levels of alanine aminotransferase (ALT) for drug-induced liver injury often assume that the biomarker is released passively from dying cells. However, the mechanisms driving that release have not been explored experimentally. The usefulness of ALT and related biomarkers will improve by developing mechanism-based explanations of elevated levels that can be expanded and elaborated incrementally. We provide the means to challenge the ability of closely related concretized model mechanisms to generate patterns of simulated hepatic injury and ALT release that scale (or not) to be quantitatively similar to the wet-lab validation targets. The validation targets for this work are elevated measures of plasma ALT following acetaminophen (APAP) exposure in mice. We build on a published model mechanism that helps explain the generation of characteristic spatiotemporal features of APAP hepatotoxicity within hepatic lobules. Discrete event and agent-oriented software methods are most prominent. We instantiate and leverage a small constellation of concrete model mechanisms. Their details during execution help bring into focus ways in which particular sources of uncertainty become entangled within and across several levels with cause-effect details. Monte Carlo sampling and simulations comprise a virtual experiment. Falsification of one (or more) of the model mechanisms provides new knowledge and shrinks the model mechanism constellation incrementally. We challenge the sufficiency of four potentially explanatory theories for ALT release. The first model mechanism tested failed to achieve the initial validation target, but each of the three others succeeded. We scale ALT amounts in virtual mice directly to target plasma ALT measures in individual mice. Results for one of the three model mechanisms matched all target ALT measures quantitatively. We assert that the actual mechanisms responsible for ALT measures in individual mice and the virtual causal processes occurring during model execution are strongly analogous within and among real hepatic lobular levels.Author summary: Interpretations of elevated biomarkers for drug-induced liver injury assume passive release during hepatocyte death, yet indirect evidence indicates that plasma levels can increase absent injury. Limitations on measurements make it infeasible to resolve causal linkages between drug disposition and plasma levels of biomarkers. To improve explanatory knowledge, we instantiate within virtual mice, plausible mechanism-based causal linkages between acetaminophen disposition and alanine aminotransferase (ALT) behavior that enables simulation results to meet stringent quantitative validation prerequisites. We challenge the sufficiency of four model mechanisms by scaling ALT measurements in virtual mice to corresponding plasma values. Virtual experiment results in which ALT release is a combined consequence of lobular-location-dependent hepatocyte death and drug-induced cellular damage, matches all validation targets. We assert that the actual mechanisms responsible for plasma ALT measures in individual mice and the virtual causal processes occurring during model execution are strongly analogous within and among real hepatic lobular levels.