Learning dynamical models of single and collective cell migration: a review

Single and collective cell migration are fundamental processes critical for physiological phenomena ranging from embryonic development and immune response to wound healing and cancer metastasis. To understand cell migration from a physical perspective, a broad variety of models for the underlying physical mechanisms that govern cell motility have been developed. A key challenge in the development of such models is how to connect them to experimental observations, which often exhibit complex stochastic behaviours. In this review, we discuss recent advances in data-driven theoretical approaches that directly connect with experimental data to infer dynamical models of stochastic cell migration. Leveraging advances in nanofabrication, image analysis, and tracking technology, experimental studies now provide unprecedented large datasets on cellular dynamics. In parallel, theoretical efforts have been directed towards integrating such datasets into physical models from the single cell to the tissue scale with the aim of conceptualising the emergent behaviour of cells. We first review how this inference problem has been addressed in both freely migrating and confined cells. Next, we discuss why these dynamics typically take the form of underdamped stochastic equations of motion, and how such equations can be inferred from data. We then review applications of data-driven inference and machine learning approaches to heterogeneity in cell behaviour, subcellular degrees of freedom, and to the collective dynamics of multicellular systems. Across these applications, we emphasise how data-driven methods can be integrated with physical active matter models of migrating cells, and help reveal how underlying molecular mechanisms control cell behaviour. Together, these data-driven approaches are a promising avenue for building physical models of cell migration directly from experimental data, and for providing conceptual links between different length-scales of description.

Data-driven Stochastic Model for Quantifying the Interplay Between Amyloid-beta and Calcium Levels in Alzheimer's Disease

The abnormal aggregation of extracellular amyloid-β ( A β ) in senile plaques resulting in calcium C a + 2 dyshomeostasis is one of the primary symptoms of Alzheimer's disease (AD). Significant research efforts have been devoted in the past to better understand the underlying molecular mechanisms driving A β deposition and C a + 2 dysregulation. Importantly, synaptic impairments, neuronal loss, and cognitive failure in AD patients are all related to the buildup of intraneuronal A β accumulation. Moreover, increasing evidence show a feed-forward loop between A β and C a + 2 levels, i.e. A β disrupts neuronal C a + 2 levels, which in turn affects the formation of A β . To better understand this interaction, we report a novel stochastic model where we analyze the positive feedback loop between A β and C a + 2 using ADNI data. A good therapeutic treatment plan for AD requires precise predictions. Stochastic models offer an appropriate framework for modelling AD since AD studies are observational in nature and involve regular patient visits. The etiology of AD may be described as a multi-state disease process using the approximate Bayesian computation method. So, utilizing ADNI data from 2-year visits for AD patients, we employ this method to investigate the interplay between A β and C a + 2 levels at various disease development phases. Incorporating the ADNI data in our physics-based Bayesian model, we discovered that a sufficiently large disruption in either A β metabolism or intracellular C a + 2 homeostasis causes the relative growth rate in both C a + 2 and A β , which corresponds to the development of AD. The imbalance of C a + 2 ions causes A β disorders by directly or indirectly affecting a variety of cellular and subcellular processes, and the altered homeostasis may worsen the abnormalities of C a + 2 ion transportation and deposition. This suggests that altering the C a + 2 balance or the balance between A β and C a + 2 by chelating them may be able to reduce disorders associated with AD and open up new research possibilities for AD therapy.

Know Before You Go: Data-Driven Beach Water Quality Forecasting

Forecasting environmental hazards is critical in preventing or building resilience to their impacts on human communities and ecosystems. Environmental data science is an emerging field that can be harnessed for forecasting, yet more work is needed to develop methodologies that can leverage increasingly large and complex data sets for decision support. Here, we design a data-driven framework that can, for the first time, forecast bacterial standard exceedances at marine beaches with 3 days lead time. Using historical data sets collected at two California sites, we train nearly 400 forecast models using statistical and machine learning techniques and test forecasts against predictions from both a naive "persistence" model and a baseline nowcast model. Overall, forecast models are found to have similar sensitivities and specificities to the persistence model, but significantly higher areas under the ROC curve (a metric distinguishing a model's ability to effectively parse classes across decision thresholds), suggesting that forecasts can provide enhanced information beyond past observations alone. Forecast model performance at all lead times was similar to that of nowcast models. Together, results suggest that integrating the forecasting framework developed in this study into beach management programs can enable better public notification and aid in proactive pollution and health risk management.

Data-driven estimation of nitric oxide emissions from global soils based on dominant vegetation covers

Soils are a major source of global nitric oxide (NO) emissions. However, estimates of soil NO emissions have large uncertainties due to limited observations and multifactorial impacts. Here, we mapped global soil NO emissions, integrating 1356 in-situ NO observations from globally distributed sites with high-resolution climate, soil, and management practice data. We then calculated global and national total NO budgets and revealed the contributions of cropland, grassland, and forest to global soil NO emissions at the national level. The results showed that soil NO emissions were explained mainly by N input, water input and soil pH. Total above-soil NO emissions of the three vegetation cover types were 9.4 Tg N year-1 in 2014, including 5.9 Tg N year-1 (1.04, 95% confidence interval [95% CI]: 0.09-1.99 kg N ha-1  year-1 ) emitted from forest, 1.7 Tg N year-1 (0.68, 95% CI: 0.10-1.26 kg N ha-1  year-1 ) from grassland, and 1.8 Tg N year-1 (0.98, 95% CI: 0.42-1.53 kg N ha-1  year-1 ) from cropland. Soil NO emissions in approximately 57% of 213 countries surveyed were dominated by forests. Our results provide updated inventories of global and national soil NO emissions based on robust data-driven models. These estimates are critical to guiding the mitigation of soil NO emissions and can be used in combination with biogeochemical models.

Big Data and Machine Learning to Improve European Grapevine Moth (Lobesia botrana) Predictions

Machine Learning (ML) techniques can be used to convert Big Data into valuable information for agri-environmental applications, such as predictive pest modeling. Lobesia botrana (Denis & Schiffermüller) 1775 (Lepidoptera: Tortricidae) is one of the main pests of grapevine, causing high productivity losses in some vineyards worldwide. This work focuses on the optimization of the Touzeau model, a classical correlation model between temperature and L. botrana development using data-driven models. Data collected from field observations were combined with 30 GB of registered weather data updated every 30 min to train the ML models and make predictions on this pest's flights, as well as to assess the accuracy of both Touzeau and ML models. The results obtained highlight a much higher F1 score of the ML models in comparison with the Touzeau model. The best-performing model was an artificial neural network of four layers, which considered several variables together and not only the temperature, taking advantage of the ability of ML models to find relationships in nonlinear systems. Despite the room for improvement of artificial intelligence-based models, the process and results presented herein highlight the benefits of ML applied to agricultural pest management strategies.

Modeling of Hydrodynamic Cavitation Reactors: Reflections on Present Status and Path Forward

Hydrodynamic cavitation (HC) is finding ever increasing applications in water, energy, chemicals, and materials sectors. HC generates intense shear, localized hot spots, and hydroxyl radicals, which are harnessed for realizing desired physicochemical transformations. Despite identification of HC as one of the most promising technology platforms, its potential is not yet adequately translated in practice. Lack of appropriate models for design, optimization, and scale-up of HC reactors is one of the primary reasons for this. In this work, the current status of modeling of HC reactors is presented. Various prevailing approaches covering empirical, phenomenological, and multiscale models are critically reviewed in light of personal experience of their application. Use of these approaches for different applications such as biomass pretreatment and wastewater treatment is briefly discussed. Some comments on extending these models for other applications like emulsions and crystallization are included. The presented models and discussion will be useful for practicing engineers and scientists interested in applying HC for a variety of applications. Some thoughts on further advances in modeling of HC reactors and outlook are shared, which may stimulate further research on improving the fidelity of computational models of HC reactors.

Predicting Van der Waals Heterostructures by a Combined Machine Learning and Density Functional Theory Approach

Van der Waals (vdW) heterostructures are constructed by different two-dimensional (2D) monolayers vertically stacked and weakly coupled by van der Waals interactions. VdW heterostructures often possess rich physical and chemical properties that are unique to their constituent monolayers. As many 2D materials have been recently identified, the combinatorial configuration space of vdW-stacked heterostructures grows exceedingly large, making it difficult to explore through traditional experimental or computational approaches in a trial-and-error manner. Here, we present a computational framework that combines first-principles electronic structure calculations, 2D material database, and supervised machine learning methods to construct efficient data-driven models capable of predicting electronic and structural properties of vdW heterostructures from their constituent monolayer properties. We apply this approach to predict the band gap, band edges, interlayer distance, and interlayer binding energy of vdW heterostructures. Our data-driven model will open avenues for efficient screening and discovery of low-dimensional vdW heterostructures and moiré superlattices with desired electronic and optical properties for targeted device applications.

Machine-Learning-Based Genome-Wide Association Studies for Uncovering QTL Underlying Soybean Yield and Its Components

A genome-wide association study (GWAS) is currently one of the most recommended approaches for discovering marker-trait associations (MTAs) for complex traits in plant species. Insufficient statistical power is a limiting factor, especially in narrow genetic basis species, that conventional GWAS methods are suffering from. Using sophisticated mathematical methods such as machine learning (ML) algorithms may address this issue and advance the implication of this valuable genetic method in applied plant-breeding programs. In this study, we evaluated the potential use of two ML algorithms, support-vector machine (SVR) and random forest (RF), in a GWAS and compared them with two conventional methods of mixed linear models (MLM) and fixed and random model circulating probability unification (FarmCPU), for identifying MTAs for soybean-yield components. In this study, important soybean-yield component traits, including the number of reproductive nodes (RNP), non-reproductive nodes (NRNP), total nodes (NP), and total pods (PP) per plant along with yield and maturity, were assessed using a panel of 227 soybean genotypes evaluated at two locations over two years (four environments). Using the SVR-mediated GWAS method, we were able to discover MTAs colocalized with previously reported quantitative trait loci (QTL) with potential causal effects on the target traits, supported by the functional annotation of candidate gene analyses. This study demonstrated the potential benefit of using sophisticated mathematical approaches, such as SVR, in a GWAS to complement conventional GWAS methods for identifying MTAs that can improve the efficiency of genomic-based soybean-breeding programs.

Epistatic models predict mutable sites in SARS-CoV-2 proteins and epitopes

During the COVID pandemic, new severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variants emerge and spread, some being of major concern due to their increased infectivity or capacity to reduce vaccine efficiency. Anticipating mutations, which might give rise to new variants, would be of great interest. We construct sequence models predicting how mutable SARS-CoV-2 positions are, using a single SARS-CoV-2 sequence and databases of other coronaviruses. Predictions are tested against available mutagenesis data and the observed variability of SARS-CoV-2 proteins. Interestingly, predictions agree increasingly with observations, as more SARS-CoV-2 sequences become available. Combining predictions with immunological data, we find an overrepresentation of mutations in current variants of concern. The approach may become relevant for potential outbreaks of future viral diseases.

The emergence of new variants of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is a major concern given their potential impact on the transmissibility and pathogenicity of the virus as well as the efficacy of therapeutic interventions. Here, we predict the mutability of all positions in SARS-CoV-2 protein domains to forecast the appearance of unseen variants. Using sequence data from other coronaviruses, preexisting to SARS-CoV-2, we build statistical models that not only capture amino acid conservation but also more complex patterns resulting from epistasis. We show that these models are notably superior to conservation profiles in estimating the already observable SARS-CoV-2 variability. In the receptor binding domain of the spike protein, we observe that the predicted mutability correlates well with experimental measures of protein stability and that both are reliable mutability predictors (receiver operating characteristic areas under the curve ∼0.8). Most interestingly, we observe an increasing agreement between our model and the observed variability as more data become available over time, proving the anticipatory capacity of our model. When combined with data concerning the immune response, our approach identifies positions where current variants of concern are highly overrepresented. These results could assist studies on viral evolution and future viral outbreaks and, in particular, guide the exploration and anticipation of potentially harmful future SARS-CoV-2 variants.

Modeling sequence-space exploration and emergence of epistatic signals in protein evolution

During their evolution, proteins explore sequence space via an interplay between random mutations and phenotypic selection. Here, we build upon recent progress in reconstructing data-driven fitness landscapes for families of homologous proteins, to propose stochastic models of experimental protein evolution. These models predict quantitatively important features of experimentally evolved sequence libraries, like fitness distributions and position-specific mutational spectra. They also allow us to efficiently simulate sequence libraries for a vast array of combinations of experimental parameters like sequence divergence, selection strength, and library size. We showcase the potential of the approach in reanalyzing two recent experiments to determine protein structure from signals of epistasis emerging in experimental sequence libraries. To be detectable, these signals require sufficiently large and sufficiently diverged libraries. Our modeling framework offers a quantitative explanation for different outcomes of recently published experiments. Furthermore, we can forecast the outcome of time- and resource-intensive evolution experiments, opening thereby a way to computationally optimize experimental protocols.

Data-Driven Modeling of the Cellular Pharmacokinetics of Degradable Chitosan-Based Nanoparticles

Nanoparticle drug delivery vehicles introduce multiple pharmacokinetic processes, with the delivery, accumulation, and stability of the therapeutic molecule influenced by nanoscale processes. Therefore, considering the complexity of the multiple interactions, the use of data-driven models has critical importance in understanding the interplay between controlling processes. We demonstrate data simulation techniques to reproduce the time-dependent dose of trimethyl chitosan nanoparticles in an ND7/23 neuronal cell line, used as an in vitro model of native peripheral sensory neurons. Derived analytical expressions of the mean dose per cell accurately capture the pharmacokinetics by including a declining delivery rate and an intracellular particle degradation process. Comparison with experiment indicates a supply time constant, τ = 2 h. and a degradation rate constant, b = 0.71 h⁻¹. Modeling the dose heterogeneity uses simulated data distributions, with time dependence incorporated by transforming data-bin values. The simulations mimic the dynamic nature of cell-to-cell dose variation and explain the observed trend of increasing numbers of high-dose cells at early time points, followed by a shift in distribution peak to lower dose between 4 to 8 h and a static dose profile beyond 8 h.

Towards Providing Effective Data-Driven Responses to Predict the Covid-19 in São Paulo and Brazil

São Paulo is the most populous state in Brazil, home to around 22% of the country's population. The total number of Covid-19-infected people in São Paulo has reached more than 1 million, while its total death toll stands at 25% of all the country's fatalities. Joining the Brazilian academia efforts in the fight against Covid-19, in this paper we describe a unified framework for monitoring and forecasting the Covid-19 progress in the state of São Paulo. More specifically, a freely available, online platform to collect and exploit Covid-19 time-series data is presented, supporting decision-makers while still allowing the general public to interact with data from different regions of the state. Moreover, a novel forecasting data-driven method has also been proposed, by combining the so-called Susceptible-Infectious-Recovered-Deceased model with machine learning strategies to better fit the mathematical model's coefficients for predicting Infections, Recoveries, Deaths, and Viral Reproduction Numbers. We show that the obtained predictor is capable of dealing with badly conditioned data samples while still delivering accurate 10-day predictions. Our integrated computational system can be used for guiding government actions mainly in two basic aspects: real-time data assessment and dynamic predictions of Covid-19 curves for different regions of the state. We extend our analysis and investigation to inspect the virus spreading in Brazil in its regions. Finally, experiments involving the Covid-19 advance in other countries are also given.

Computational medicine, present and the future: obstetrics and gynecology perspective

Medicine is, in its essence, decision making under uncertainty; the decisions are made about tests to be performed and treatments to be administered. Traditionally, the uncertainty in decision making was handled using expertise collected by individual providers and, more recently, systematic appraisal of research in the form of evidence-based medicine. The traditional approach has been used successfully in medicine for a very long time. However, it has substantial limitations because of the complexity of the system of the human body and healthcare. The complex systems are a network of highly coupled components intensely interacting with each other. These interactions give those systems redundancy and thus robustness to failure and, at the same time, equifinality, that is, many different causative pathways leading to the same outcome. The equifinality of the complex systems of the human body and healthcare system demand the individualization of medical care, medicine, and medical decision making. Computational models excel in modeling complex systems and, consequently, enabling individualization of medical decision making and medicine. Computational models are theory- or knowledge-based models, data-driven models, or models that combine both approaches. Data are essential, although to a different degree, for computational models to successfully represent complex systems. The individualized decision making, made possible by the computational modeling of complex systems, has the potential to revolutionize the entire spectrum of medicine from individual patient care to policymaking. This approach allows applying tests and treatments to individuals who receive a net benefit from them, for whom benefits outweigh the risk, rather than treating all individuals in a population because, on average, the population benefits. Thus, the computational modeling-enabled individualization of medical decision making has the potential to both improve health outcomes and decrease the costs of healthcare.

Electroencephalography-Derived Prognosis of Functional Recovery in Acute Stroke Through Machine Learning Approaches

Stroke, if not lethal, is a primary cause of disability. Early assessment of markers of recovery can allow personalized interventions; however, it is difficult to deliver indexes in the acute phase able to predict recovery. In this perspective, evaluation of electrical brain activity may provide useful information. A machine learning approach was explored here to predict post-stroke recovery relying on multi-channel electroencephalographic (EEG) recordings of few minutes performed at rest. A data-driven model, based on partial least square (PLS) regression, was trained on 19-channel EEG recordings performed within 10 days after mono-hemispheric stroke in 101 patients. The band-wise (delta: 1-4[Formula: see text]Hz, theta: 4-7[Formula: see text]Hz, alpha: 8-14[Formula: see text]Hz and beta: 15-30[Formula: see text]Hz) EEG effective powers were used as features to predict the recovery at 6 months (based on clinical status evaluated through the NIH Stroke Scale, NIHSS) in an optimized and cross-validated framework. In order to exploit the multimodal contribution to prognosis, the EEG-based prediction of recovery was combined with NIHSS scores in the acute phase and both were fed to a nonlinear support vector regressor (SVR). The prediction performance of EEG was at least as good as that of the acute clinical status scores. A posteriori evaluation of the features exploited by the analysis highlighted a lower delta and higher alpha activity in patients showing a positive outcome, independently of the affected hemisphere. The multimodal approach showed better prediction capabilities compared to the acute NIHSS scores alone ([Formula: see text] versus [Formula: see text], AUC = 0.80 versus AUC = 0.70, [Formula: see text]). The multimodal and multivariate model can be used in acute phase to infer recovery relying on standard EEG recordings of few minutes performed at rest together with clinical assessment, to be exploited for early and personalized therapies. The easiness of performing EEG may allow such an approach to become a standard-of-care and, thanks to the increasing number of labeled samples, further improving the model predictive power.

Data science in neurodegenerative disease: its capabilities, limitations, and perspectives

PURPOSE OF REVIEW

With the advancement of computational approaches and abundance of biomedical data, a broad range of neurodegenerative disease models have been developed. In this review, we argue that computational models can be both relevant and useful in neurodegenerative disease research and although the current established models have limitations in clinical practice, artificial intelligence has the potential to overcome deficiencies encountered by these models, which in turn can improve our understanding of disease.

RECENT FINDINGS

In recent years, diverse computational approaches have been used to shed light on different aspects of neurodegenerative disease models. For example, linear and nonlinear mixed models, self-modeling regression, differential equation models, and event-based models have been applied to provide a better understanding of disease progression patterns and biomarker trajectories. Additionally, the Cox-regression technique, Bayesian network models, and deep-learning-based approaches have been used to predict the probability of future incidence of disease, whereas nonnegative matrix factorization, nonhierarchical cluster analysis, hierarchical agglomerative clustering, and deep-learning-based approaches have been employed to stratify patients based on their disease subtypes. Furthermore, the interpretation of neurodegenerative disease data is possible through knowledge-based models which use prior knowledge to complement data-driven analyses. These knowledge-based models can include pathway-centric approaches to establish pathways perturbed in a given condition, as well as disease-specific knowledge maps, which elucidate the mechanisms involved in a given disease. Collectively, these established models have revealed high granular details and insights into neurodegenerative disease models.

SUMMARY

In conjunction with increasingly advanced computational approaches, a wide spectrum of neurodegenerative disease models, which can be broadly categorized into data-driven and knowledge-driven, have been developed. We review the state of the art data and knowledge-driven models and discuss the necessary steps which are vital to bring them into clinical application.

Modelling perceptions of criminality and remorse from faces using a data-driven computational approach

Perceptions of criminality and remorse are critical for legal decision-making. While faces perceived as criminal are more likely to be selected in police lineups and to receive guilty verdicts, faces perceived as remorseful are more likely to receive less severe punishment recommendations. To identify the information that makes a face appear criminal and/or remorseful, we successfully used two different data-driven computational approaches that led to convergent findings: one relying on the use of computer-generated faces, and the other on photographs of people. In addition to visualising and validating the perceived looks of criminality and remorse, we report correlations with earlier face models of dominance, threat, trustworthiness, masculinity/femininity, and sadness. The new face models of criminal and remorseful appearance contribute to our understanding of perceived criminality and remorse. They can be used to study the effects of perceived criminality and remorse on decision-making; research that can ultimately inform legal policies.

Modelling the propagation of social response during a disease outbreak

Epidemic trajectories and associated social responses vary widely between populations, with severe reactions sometimes observed. When confronted with fatal or novel pathogens, people exhibit a variety of behaviours from anxiety to hoarding of medical supplies, overwhelming medical infrastructure and rioting. We developed a coupled network approach to understanding and predicting social response. We couple the disease spread and panic spread processes and model them through local interactions between agents. The social contagion process depends on the prevalence of the disease, its perceived risk and a global media signal. We verify the model by analysing the spread of disease and social response during the 2009 H1N1 outbreak in Mexico City and 2003 severe acute respiratory syndrome and 2009 H1N1 outbreaks in Hong Kong, accurately predicting population-level behaviour. This kind of empirically validated model is critical to exploring strategies for public health intervention, increasing our ability to anticipate the response to infectious disease outbreaks.

Constructing predictive models of human running

Running is an essential mode of human locomotion, during which ballistic aerial phases alternate with phases when a single foot contacts the ground. The spring-loaded inverted pendulum (SLIP) provides a starting point for modelling running, and generates ground reaction forces that resemble those of the centre of mass (CoM) of a human runner. Here, we show that while SLIP reproduces within-step kinematics of the CoM in three dimensions, it fails to reproduce stability and predict future motions. We construct SLIP control models using data-driven Floquet analysis, and show how these models may be used to obtain predictive models of human running with six additional states comprising the position and velocity of the swing-leg ankle. Our methods are general, and may be applied to any rhythmic physical system. We provide an approach for identifying an event-driven linear controller that approximates an observed stabilization strategy, and for producing a reduced-state model which closely recovers the observed dynamics.

Stimulus dependence of local field potential spectra: experiment versus theory

The local field potential (LFP) captures different neural processes, including integrative synaptic dynamics that cannot be observed by measuring only the spiking activity of small populations. Therefore, investigating how LFP power is modulated by external stimuli can offer important insights into sensory neural representations. However, gaining such insight requires developing data-driven computational models that can identify and disambiguate the neural contributions to the LFP. Here, we investigated how networks of excitatory and inhibitory integrate-and-fire neurons responding to time-dependent inputs can be used to interpret sensory modulations of LFP spectra. We computed analytically from such models the LFP spectra and the information that they convey about input and used these analytical expressions to fit the model to LFPs recorded in V1 of anesthetized macaques (Macaca mulatta) during the presentation of color movies. Our expressions explain 60%-98% of the variance of the LFP spectrum shape and its dependency upon movie scenes and we achieved this with realistic values for the best-fit parameters. In particular, synaptic best-fit parameters were compatible with experimental measurements and the predictions of firing rates, based only on the fit of LFP data, correlated with the multiunit spike rate recorded from the same location. Moreover, the parameters characterizing the input to the network across different movie scenes correlated with cross-scene changes of several image features. Our findings suggest that analytical descriptions of spiking neuron networks may become a crucial tool for the interpretation of field recordings.

Individualized Prediction of Heat Stress in Firefighters: A Data-Driven Approach Using Classification and Regression Trees

The purpose of this study was to explore data-driven models, based on decision trees, to develop practical and easy to use predictive models for early identification of firefighters who are likely to cross the threshold of hyperthermia during live-fire training. Predictive models were created for three consecutive live-fire training scenarios. The final predicted outcome was a categorical variable: will a firefighter cross the upper threshold of hyperthermia - Yes/No. Two tiers of models were built, one with and one without taking into account the outcome (whether a firefighter crossed hyperthermia or not) from the previous training scenario. First tier of models included age, baseline heart rate and core body temperature, body mass index, and duration of training scenario as predictors. The second tier of models included the outcome of the previous scenario in the prediction space, in addition to all the predictors from the first tier of models. Classification and regression trees were used independently for prediction. The response variable for the regression tree was the quantitative variable: core body temperature at the end of each scenario. The predicted quantitative variable from regression trees was compared to the upper threshold of hyperthermia (38°C) to predict whether a firefighter would enter hyperthermia. The performance of classification and regression tree models was satisfactory for the second (success rate = 79%) and third (success rate = 89%) training scenarios but not for the first (success rate = 43%). Data-driven models based on decision trees can be a useful tool for predicting physiological response without modeling the underlying physiological systems. Early prediction of heat stress coupled with proactive interventions, such as pre-cooling, can help reduce heat stress in firefighters.