PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Model interpretability”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 109 records · Page 6Linked to original sources
PubMed2026

Enterocutaneous Fistula-Associated Sepsis and Mortality: Development and Validation of a Multimodal Artificial Intelligence Prediction Model.

BACKGROUND: Predicting enterocutaneous fistula (ECF)-associated sepsis and mortality poses significant challenges in digital health care due to the disease's complexity and heterogeneous clinical manifestations. Current approaches that rely on single-modal data or traditional scoring systems often fail to capture the intricate immune-inflammatory dynamics and multisystem involvement in patients with ECF. OBJECTIVE: This study aims to develop an artificial intelligence (AI)-driven multimodal fusion model integrating clinical, imaging, and transcriptomic data for early prediction of ECF-associated sepsis and 28-day mortality, addressing the limitations of conventional single-dimensional models. METHODS: This study leveraged publicly available datasets (Medical Information Mart for Intensive Care III [MIMIC-III], electronic Intensive Care Unit [eICU], and The Cancer Genome Atlas) to construct a multimodal framework. Clinical parameters were processed using Extreme Gradient Boosting, abdominal imaging features were extracted via convolutional neural networks, and transcriptomic profiles were analyzed with variational autoencoders. A Transformer-based fusion network was employed for joint prediction and validated through cross-validation and external testing. Key features were identified using Shapley Additive Explanations and Local Interpretable Model-Agnostic Explanations interpretability algorithms, while immune regulatory mechanisms were explored via weighted gene co-expression network analysis. RESULTS: The multimodal model achieved an area under the curve (AUC) of 0.89 for predicting sepsis and 28-day mortality, outperforming unimodal models (clinical-only model, AUC 0.72, and imaging-only model, AUC 0.78). Critical predictors included Sequential Organ Failure Assessment score, lactate levels, intra-abdominal free fluid on imaging, and immunoregulatory genes (programmed death-ligand 1 [PD-L1] and indoleamine 2,3-dioxygenase 1 [IDO1]). Mechanistic analysis revealed distinct immune reprogramming in patients with sepsis, characterized by increased regulatory T cells and M2 macrophages, along with downregulated cluster of differentiation 8+ (CD8+) T cells. CONCLUSIONS: This multimodal AI model offers an innovative digital solution in medical informatics, enabling precise early risk stratification for ECF-associated sepsis. By integrating multisource data and providing interpretable insights into immune-inflammatory pathways, the model enhances health care quality for patients with ECF and paves the way for personalized intervention strategies.

Humans↗
PubMed1990

An empirical model for analysing and interpreting ventricular measures.

Given the abundant and, at times, contradictory studies of ventricular enlargement in neurological and psychiatric disorders, the current study was carried out to provide an empirical basis for analysing and interpreting these measures. A sample of CT scans on 100 control subject was drawn from the files of the University of British Columbia Department of Radiology and 19 measures of ventricular and head diameter or area were made. The interrelationships of these measures were then examined using factor analytic procedures. Three ventricular dimensions were found. To validate these three dimensions, the relationship of each with age was examined and then the age-corrected scores of seven clinical groups were compared on each of these dimensions. The magnitude of the relationship between these dimensions and age was impressive and each dimension contributed unique information regarding these age-related changes. Moreover, the analysis of the clinical groups suggested that differential patterns of ventricular change were present dependent upon the disease. These results are discussed with a view to integrating the findings of previous studies and planning future studies.

Adult↗
PubMed1979

Metabolism of totally ischemic excised dog heart. II. Interpretation of a computer model.

Analysis of the ischemic dog heart preparation described in the preceding paper indicates that it is an analogue in slow motion of the tissue in the center of a cardiac infarct. It is respiring very slowly and not capable of performing mechanical work. Glycolysis starts up with both glucose and glycogen as inputs. Later hexokinase and to some extent phosphofructokinase become limiting owing to inhibitor accumulation or acidosis. Metabolism then results primarily from cAMP-driven glycogenolysis, largely limited by the glycogen debranching enzymes at later times, with accumultion not only of lactate and alpha-glycerophosphate but of glucose as well. Amino acid levels oscillate with time while fatty acids accumulate at late times. The elevation of cAMP at later times may involve disturbances in its metabolism as well as mechanisms such as adenosine accumulation that are more important in cardiac ischemia than in normal heart. The clinical implications of this behavior are discussed.

Amino Acids↗
PubMed1991

Effects of gene mutations in lipoprotein and hepatic lipases as interpreted by a molecular model of the pancreatic triglyceride lipase.

A molecular model of human pancreatic lipase (Winkler, F. K., D'Arcy, A., and Hunziker, W. (1990) Nature 343, 771-774) is used to explain the possible structural effects of the amino acid mutations identified to date in the human lipoprotein and hepatic lipase genes. A sequence homology profile was used to evaluate the alignment of the amino acid sequences of all three lipolytic enzymes (Kirchgessner, T. G., Chuat, J.-C., Heinzmann, C., Etienne, J., Guilhot, S., Svenson, K., Ameis, D., Pilon, C., D'Auriol, L., Andalibi, A., Schotz, M. C., Galibert, F., and Lusis, A. J. (1989) Proc. Natl. Acad. Sci. U. S. A. 86, 9647-9651) with respect to the secondary structure elements identified in the pancreatic lipase. As expected, maximum homology is observed in internal regions namely the hydrophobic strands of the central beta-pleated sheet. This observation strongly supports the hypothesis that all three molecules exhibit a very similar three-dimensional structure, particularly in the N-terminal catalytic domain. There is considerable variation in some of the surface loops connecting the individual strands, whereas others are conserved. It is hypothesized that the most conserved loops located around the active site are responsible for the catalytic function (similar for all three enzymes), whereas those that markedly differ are involved in the regulation at the molecular level, namely the binding of colipase (pancreatic enzyme) and apolipoprotein CII (lipoprotein lipase). The currently available library of hepatic and lipoprotein gene mutations seems to indicate that the majority of mutants disrupt the folding of the polypeptide chain, rather than affect specific constellations in and around the catalytic site or regulatory loops.

Amino Acid Sequence↗
PubMed1990

Uncertainties in pharmacokinetic modeling for perchloroethylene. I. Comparison of model structure, parameters, and predictions for low-dose metabolism rates for models derived by different authors.

In recent years physiologically based pharmacokinetic models have come to play an increasingly important role in risk assessment for carcinogens. The hope is that they can help open the black box between external exposure and carcinogenic effects to experimental observations, and improve both high-dose to low-dose and interspecies projections of risk. However, to date, there have been only relatively preliminary efforts to assess the uncertainties in current modeling results. In this paper we compare the physiologically based pharmacokinetic models (and model predictions of risk-related overall metabolism) that have been produced by seven different sets of authors for perchloroethylene (tetrachloroethylene). The most striking conclusion from the data is that most of the differences in risk-related model predictions are attributable to the choice of the data sets used for calibrating the metabolic parameters. Second, it is clear that the bottom-line differences among the model predictions are appreciable. Overall, the ratios of low-dose human to bioassay rodent metabolism spanned a 30-fold range for the six available human/rat comparisons, and the seven predicted ratios of low-dose human to bioassay mouse metabolism spanned a 13-fold range. (The greater range for the rat/human comparison is attributable to a structural assumption by one author group of competing linear and saturable pathways, and their conclusion that the dangerous saturable pathway constitutes a minor fraction of metabolism in rats.) It is clear that there are a number of opportunities for modelers to make different choices of model structure, interpretive assumptions, and calibrating data in the process of constructing pharmacokinetic models for use in estimating "delivered" or "biologically effective" dose for carcinogenesis risk assessments. We believe that in presenting the results of such modeling studies, it is important for researchers to explore the results of alternative, reasonably likely approaches for interpreting the available data--and either show that any conclusions they make are relatively insensitive to particular interpretive choices, or to acknowledge the differences in conclusions that would result from plausible alternative views of the world.

Animals↗
PubMed1977

[Electronic measuring and calculating devices for arcogrammetric model diagnosis and for the interpretation of teleradiographs].

A newly developed method to mesure different parameters from plaster models of the teeth, from dental radiographs and from cephalometric X-rays by means of a 4-K minicomputer and on-line linear transducers are described. Programs in connection to Arcogrammetrics [Herren] are presented. The electronic devices allow storage of these parameters in order to make drawings of the actual dental arches and of predicted arches, as well as to trace growth and/or progress in orthodontic treatment.

Cephalometry↗
PubMed1989

Variability of safe dose estimates when using complicated models of the carcinogenic process. A case study: methylene chloride.

Advances in understanding carcinogenesis have led to the development of mathematical models that have biologically interpretable parameters. These models utilize more of the available scientific data than the empirical models routinely employed for quantifying carcinogenic risk. They also require consideration of sources of uncertainty in risk estimates that were previously ignored, such as animal-to-animal variability of physiological and pharmacological constants. A numerical technique is proposed for studying the consequences of incorporating the intrapopulation variability of biologically interpretable parameters into the risk assessment process. To demonstrate the technique, the variability of safe dose estimates for exposure to methylene chloride is considered. The results suggest that intrapopulation variability of the model parameters can increase the variability of safe dose estimates an appreciable amount.

Animals↗
PubMed2026

Assessing Metal Ion Assignment Accuracy in Protein Data Bank Models via Elemental Spectroscopy.

Accurate representation of metal ions in macromolecular structures is critical for chemical interpretation, computational modeling, and machine-learning methods that rely on Protein Data Bank (PDB) entries. However, the elemental identity of metals modeled in crystallographic structures is often inferred indirectly and rarely validated experimentally. Here, we combine Particle Induced X-ray Emission (PIXE) and X-ray Fluorescence Spectroscopy (XRFS) to determine the elemental composition of protein samples used to generate 70 deposited metalloprotein crystal structures. By analyzing the original protein material employed for crystallization, but before the addition of crystallization buffer solutions, we assess whether the modeled metal ions in deposited structures are consistent with experimentally detectable elemental content. We find that in a majority of cases, the metals modeled in the corresponding PDB entries are inconsistent with the metals present in the protein samples before crystallization, or that additional metals are present but not represented in the structural models. Spectroscopic results were integrated with automated crystallographic validation metrics, including real-space Z-difference (RSZD) analysis and systematic rerefinement, to evaluate atomic-number mismatch at metal sites. PIXE and XRFS show strong agreement for dominant elemental signals and provide complementary, scalable approaches for identifying suspect metal assignments. This work does not address physiological or functional metalation but instead highlights a widespread data integrity issue in deposited macromolecular structures, PDB-wide. These results establish an experimentally corroborated link between elemental identity and crystallographic validation metrics, enabling the large-scale detection of chemically inconsistent annotations in structural databases used for computational modeling and machine learning.

Databases, Protein↗
PubMed1991

Analysis of failure time data with ordinal categories of response.

When failure times are observed, additional information concerning the type of failure is often recorded. A method which simultaneously models the failure times and additional information in the form of ordinal categories is discussed. An application to clinical trial data, in which the failure times are times of onset of headache, and the headaches are classified into the ordinal categories mild, moderate and severe, illustrates how this method may be used and how the final model can be interpreted. The continuation ratio model, which is used in this method, is described in detail.

Clinical Trials as Topic↗
PubMed1990

[Strategies in the use of isolated cardiomyocytes in relation to other models in experimental cardiology].

Selection of the optimal model for a specific experiment considerably determines the results and their correct interpretation. The model of the isolated cardiomyocyte is increasingly being used in experimental cardiology as it provides several advantages in comparison to models in which the heart tissue remains relatively complete. Similarly as in other models, the factors limiting its use have to be known also in the case of the isolated cardiomyocyte. To minimalize misinterpretation of results the given problem is to be handled at all available levels.

Heart↗
PubMed1991

Utility of phenomenological models for describing temperature dependence of bacterial growth.

We compared three unstructured mathematical models, the master reaction, the square root, and the damage/repair models, for describing the relationship between temperature and the specific growth rates of bacteria. The models were evaluated on the basis of several criteria: applicability, ease of use, simple interpretation of model parameters, problem-free determination of model parameters, statistical evaluation of goodness of fit (chi 2 test), and biological relevance. Best-fit parameters for the master reaction model could be obtained by using two consecutive nonlinear least-square fits. The damage/repair model proved to be unsuited for the data sets considered and was judged markedly overparameterized. The square root model allowed nonproblematical parameter estimation by a nonlinear least-square procedure and, together with the master reaction model, was able to describe the temperature dependence of the specific growth rates of Klebsiella pneumoniae NCIB 418, Escherichia coli NC3, Bacillus sp. strain NCIB 12522, and the thermotolerant coccobacillus strain NA17. The square root and master reaction models were judged to be equally valid and superior to the damage/repair model, even though the square root model is devoid of a conceptual basis.

Bacteria↗
PubMed1989

A strategy for analysing multiple risk factors with application to cervical pain syndrome.

When studying the possible effects of several factors in a given disease, two major problems arise: (1) confounding, and (2) multiplicity of tests. Frequently, in order to cope with the problem of confounding factors, models with multiple explanatory variables are used. However, the correlation structure of the variables may be such that the corresponding tests have low power: in its extreme form this situation is coined by the term "multicollinearity". As the problem of multiplicity is still relevant in these models, the interpretation of results is, in most cases, very hazardous. We propose a strategy--based on a tree structure of the variables--which provides a guide to the interpretation and controls the risk of erroneously rejecting null hypotheses. The strategy was applied to a study of cervical pain syndrome involving 990 subjects and 17 variables. Age, sex, head trauma, posture at work and psychological status were all found to be important risk factors.

Adult↗
PubMed2026

Predicting food taste with bound-driven optimization.

The prediction of sensory attributes from ingredient-level formulations is an emerging challenge at the intersection of food science and artificial intelligence. We address the fundamental question of whether the taste of a food can be predicted from its ingredients by treating recipes as composite materials. We apply Hashin-Shtrikman (HS) and Reuss-Voigt (RV) bounds, techniques originally developed for elastic moduli, as a null-hypothesis additive baseline for five taste dimensions (sweetness, sourness, bitterness, umami, saltiness) on a curated dataset of 70 recipes decomposed into 115 distinct ingredients scored against a library of 209 ingredient-level taste references with trained-panel ground truth. This baseline systematically under-predicts perceived taste: 77% of actual taste values exceeded the HS upper bound, with the exceedance rate ranging from 26% (bitterness) to 97% (saltiness). We traced this gap to specific processing chemistry (Maillard reactions, caramelization, evaporative concentration, protein hydrolysis, and nucleotide synergy) and introduced a hybrid model that augments the HS baseline with eight chemistry-proxy features encoding these mechanisms. Our results show that our interpretable hybrid model eliminates the systematic bias and reduces mean absolute error by 27%-62% for sweetness, sourness, umami, and saltiness while using only 10 interpretable features, achieving performance comparable to a black-box Lasso regression on 115 per-ingredient features. We further demonstrate constrained inverse design via Differential Evolution, recovering ingredient formulations that match target taste profiles subject to compositional bounds. Our work demonstrates how key chemical processes during food preparation can inform and augment physics-based and machine learning models, providing a quantitative fingerprint of processing chemistry's contribution to taste perception and paving the way for model-driven food formulation with targeted sensory characteristics.

Composite material bounds↗
PubMed1990

Comments on a time-dependent version of the linear-quadratic model.

The accuracy and interpretation of the "LQ + time" model (E = D(alpha + beta d) - gamma T) are discussed. Evidence is presented, based on data in the literature, that this model does not accurately describe the changes in isoeffect dose occurring with protraction of the overall treatment time during fractionated irradiation of the lung. This lack of fit of the model explains, in part, the surprisingly large values of gamma/alpha that have been derived from experimental lung data. The large apparent time factors for lung suggested by the model are also partly explained by the fact that gamma T/alpha, despite having units of dose, actually measures the influence of treatment time on the effect scale, not the dose scale, and is shown to consistently overestimate the change in total dose. The unusually high values of alpha/beta that have been derived for lung using the model (approximately 5 Gy) are shown to be influenced by the method by which the model was fitted to data. Reanalyses of the data using a more statistically valid regression procedure produce estimates of alpha/beta more typical of those usually cited for lung (approximately 3 Gy). Most importantly, published isoeffect data from lung indicate that the true deviation from the linear-quadratic (LQ) model is nonlinear in time, instead of linear, and also depends on other factors such as the effect level and the size of dose per fraction. Thus, we do not advocate the use of the "LQ + time" expression as a general isoeffect model.

Animals↗
PubMed1978

General theory of critical periods and development of obesity.

The general systems theory (GST), the general theory of organization (GTO), and the general theory of critical periods (GTCP) have been applied to some nutritional problems. This theoretical approach seems to be in good agreement with most of the data of the literature and with the personal experience, pointing at the possibility to use a simple general model to interpret the complex problem of obesity.

Critical Period, Psychological↗
PubMed1991

A framework for the interpretation of first-order interaction in logit modeling.

Several suggestions have been tendered for interpreting first-order interaction in log-linear analysis. Occasionally these methods result either in a loss of information or in results that are difficult to grasp on an intuitive level. It is argued that interpreting effect parameters in terms of odds ratios provides an elegant and intuitively appealing conceptual framework that has great generality across models. The interaction term then resembles a cross-product term in linear regression. In both forms of analysis, the partial effect of a given predictor on the response is composed of a constant and a correction that is a function of the other predictor involved in the interaction. This framework is especially appealing for models in which the logit is based on a bifurcation of the dependent variable. Although odds ratios are still useful for summarizing effects on polytomous dependent variables, greater caution must be exercised to avoid misleading interpretations.

Humans↗