COMPLETED RESEARCH
COMPLETED RESEARCH
Embedded Files
PraxisModus Institute for Advanced Studies develops hybrid mechanistic, statistical, and AI/ML modeling frameworks for infectious-disease forecasting, risk assessment, and intervention planning. In parallel, we pursue foundational research in statistical methodology for ecological scaling laws and theoretical ecology of trophic transfer. The following studies illustrate our growing research portfolio across epidemic modeling, climate-sensitive disease prediction, vector-borne disease ecology, policy-oriented decision support, statistical benchmarking, and geometric theories of trophic organization.
PraxisModus Institute for Advanced Studies develops hybrid mechanistic, statistical, and AI/ML modeling frameworks for infectious-disease forecasting, risk assessment, and intervention planning. In parallel, we pursue foundational research in statistical methodology for ecological scaling laws and theoretical ecology of trophic transfer. The following studies illustrate our growing research portfolio across epidemic modeling, climate-sensitive disease prediction, vector-borne disease ecology, policy-oriented decision support, statistical benchmarking, and geometric theories of trophic organization.
Together, the completed and ongoing studies demonstrate the PraxisModus research strategy: combining mechanistic biological understanding, advanced statistical modeling, AI/ML forecasting, intervention simulation, and foundational theoretical ecology to address real-world problems.
Together, the completed and ongoing studies demonstrate the PraxisModus research strategy: combining mechanistic biological understanding, advanced statistical modeling, AI/ML forecasting, intervention simulation, and foundational theoretical ecology to address real-world problems.
These projects form the basis for a broader PraxisModus infectious-disease modeling platform capable of integrating surveillance data, climate data, vector ecology, spatial mobility, and AI-driven forecasting, alongside a parallel effort in theoretical ecology and statistical methodology for scaling laws.
These projects form the basis for a broader PraxisModus infectious-disease modeling platform capable of integrating surveillance data, climate data, vector ecology, spatial mobility, and AI-driven forecasting, alongside a parallel effort in theoretical ecology and statistical methodology for scaling laws.
Explore Completed Research:
Explore Completed Research:
COVID-19: Dynamical Modeling of Mobility, Vaccination, and Transmission Control
COVID-19: Dynamical Modeling of Mobility, Vaccination, and Transmission Control
COVID-19: Dynamical Modeling of Mobility, Vaccination, and Transmission Control
COVID-19: Dynamical Modeling of Mobility, Vaccination, and Transmission Control
COVID-19: Dynamical Modeling of Mobility, Vaccination, and Transmission Control
COVID-19: Dynamical Modeling of Mobility, Vaccination, and Transmission Control
This study developed a time-dynamical mathematical framework to evaluate how population mobility restrictions and vaccination coverage jointly regulate the net reproductive rate of COVID-19. Using Google mobility-derived homestay indices, vaccination data, case counts, and mortality data, the study first tested alternative model hypotheses in ten island nations and then extended the analysis to 124 countries worldwide. The final model incorporated both homestay forcing and vaccination forcing, allowing the reproductive rate to be expressed as a function of mobility reduction and the proportion of the population vaccinated or recovered.
The analysis showed that the combined mobility--vaccination model explained the observed dynamics in 106 countries. Across these countries, increasing homestay percentages reduced the net reproductive rate, but the degree of mobility restriction required to bring transmission below the epidemic threshold depended strongly on vaccination and recovery levels. When vaccination-plus-recovery coverage was low, many countries required strong mobility reductions to force $R_0 < 1$. As immunity coverage increased, the required level of mobility restriction declined substantially. The work also estimated epidemiological quantities including infection fatality rate, incubation period, infectious duration, death-window timing, and effective vaccine-related reduction in susceptibility under mixed-variant conditions.
This study demonstrated how relatively simple dynamical models, when linked to real-time mobility and vaccination data, can support scenario-based epidemic control. It provided an early example of policy-oriented infectious-disease modeling that combines biological transmission structure with population-level behavioral and intervention data.
The analysis showed that the combined mobility--vaccination model explained the observed dynamics in 106 countries. Across these countries, increasing homestay percentages reduced the net reproductive rate, but the degree of mobility restriction required to bring transmission below the epidemic threshold depended strongly on vaccination and recovery levels. When vaccination-plus-recovery coverage was low, many countries required strong mobility reductions to force $R_0 < 1$. As immunity coverage increased, the required level of mobility restriction declined substantially. The work also estimated epidemiological quantities including infection fatality rate, incubation period, infectious duration, death-window timing, and effective vaccine-related reduction in susceptibility under mixed-variant conditions.
This study demonstrated how relatively simple dynamical models, when linked to real-time mobility and vaccination data, can support scenario-based epidemic control. It provided an early example of policy-oriented infectious-disease modeling that combines biological transmission structure with population-level behavioral and intervention data.
AUTHORS: M. Rajakaruna, et al.
AFFILIATIONS: Department of Mathematical and Computational Sciences, University of Toronto Mississauga, Canada; Battelle Center for Mathematical Medicine, Abigail Wexner Research Institute, Nationide Children's Hospital, Columbus, Ohio, USA; Department of Zoology, University of Peradeniya, Sri Lanka.
Rajakaruna, M., Rajakaruna, H., and Rajakaruna, R. (2022). Dynamical regulations on mobility and vaccinations for controlling COVID-19 spread. Scientific Reports, 12, 3554. https://doi.org/10.1038/s41598-022-07371-5
Dengue: Hybrid Mechanistic--AI Modeling of Urban Transmission in Sri Lanka
Dengue: Hybrid Mechanistic--AI Modeling of Urban Transmission in Sri Lanka
Dengue: Hybrid Mechanistic-AI Modeling of Urban Transmission in Sri Lanka
Dengue: Hybrid Mechanistic-AI Modeling of Urban Transmission in Sri Lanka
Dengue: Hybrid Mechanistic--AI Modeling of Urban Transmission in Sri Lanka
Dengue: Hybrid Mechanistic--AI Modeling of Urban Transmission in Sri Lanka
This study develops a hybrid mechanistic--machine learning framework to understand dengue transmission in the Colombo Municipal Council region of Sri Lanka, a hyperendemic urban setting with approximately 1.5 million daily commuters. The core scientific question is whether dengue dynamics in Colombo are driven primarily by local vector ecology or continuously sustained by infections imported from surrounding regions.
The modeling framework integrates a climate-forced SEIRS--SEI transmission model, gravity-based commuter coupling, entomological surveillance data, imputation of missing premise and container indices, maximum a posteriori parameter estimation, and machine-learning residual correction. The model was calibrated using monthly dengue, climate, entomological, and spatial incidence data from 2015 to 2025. On the independent test period, the hybrid model substantially outperformed both a standalone mechanistic model and a pure machine-learning baseline, demonstrating that biological interpretability and predictive accuracy can be combined within one operational framework.
The results indicate that dengue control in Colombo cannot be treated as a purely municipal vector-control problem. Spatial incidence from surrounding regions, including regions up to approximately 70 km away, emerged as a dominant predictor, suggesting strong regional coupling. At the same time, local vector-borne transmission exerted the strongest marginal influence on epidemic growth, meaning that imported infections may seed outbreaks while local mosquito ecology determines their amplification. This work provides a policy-relevant framework for dengue forecasting and intervention analysis in connected urban systems.
The modeling framework integrates a climate-forced SEIRS--SEI transmission model, gravity-based commuter coupling, entomological surveillance data, imputation of missing premise and container indices, maximum a posteriori parameter estimation, and machine-learning residual correction. The model was calibrated using monthly dengue, climate, entomological, and spatial incidence data from 2015 to 2025. On the independent test period, the hybrid model substantially outperformed both a standalone mechanistic model and a pure machine-learning baseline, demonstrating that biological interpretability and predictive accuracy can be combined within one operational framework.
The results indicate that dengue control in Colombo cannot be treated as a purely municipal vector-control problem. Spatial incidence from surrounding regions, including regions up to approximately 70 km away, emerged as a dominant predictor, suggesting strong regional coupling. At the same time, local vector-borne transmission exerted the strongest marginal influence on epidemic growth, meaning that imported infections may seed outbreaks while local mosquito ecology determines their amplification. This work provides a policy-relevant framework for dengue forecasting and intervention analysis in connected urban systems.
AUTHORS: M. Rajakaruna. K. Dissanayake, K. et al.
Submitted to Royal Society Open Science
Canine Anaplasmosis: Climate-Lagged Machine Learning Forecasting for One Health Surveillance
Canine Anaplasmosis: Climate-Lagged Machine Learning Forecasting for One Health Surveillance
Canine Anaplasmosis: Climate-Lagged Machine Learning Forecasting for One Health Surveillance
Canine Anaplasmosis: Climate-Lagged Machine Learning Forecasting for One Health Surveillance
Canine Anaplasmosis: Climate-Lagged Machine Learning Forecasting for One Health Surveillance
Canine Anaplasmosis: Climate-Lagged Machine Learning Forecasting for One Health Surveillance
This study investigates the environmental drivers of canine anaplasmosis in Sri Lanka using a machine-learning framework informed by tick biology and climate-lag structure. Using 132 monthly observations of canine anaplasmosis cases from veterinary clinic records in Western Province, Sri Lanka, together with rainfall and temperature data, the study compared six machine-learning approaches: Ridge regression, Lasso regression, Random Forest, Gradient Boosting, XGBoost, and Support Vector Regression. Environmental variables were lagged by one to three months to capture the delay between favorable climate conditions, tick development, and subsequent disease occurrence.
The analysis identified antecedent rainfall as a key predictor of disease incidence, with rainfall lags of one to three months showing meaningful associations with reported cases. Among the tested models, Support Vector Regression achieved the strongest predictive performance. This study provides a methodological foundation for early-warning systems for climate-sensitive tick-borne diseases in companion animals.
The analysis identified antecedent rainfall as a key predictor of disease incidence, with rainfall lags of one to three months showing meaningful associations with reported cases. Among the tested models, Support Vector Regression achieved the strongest predictive performance. This study provides a methodological foundation for early-warning systems for climate-sensitive tick-borne diseases in companion animals.
AUTHORS: M. Rajakaruna, V, Perera, et al.
Under review at a [neglected tropical diseases journal]
Statistical Methodology: Errors-in-Variables for Power-Law Scaling
Statistical Methodology: Errors-in-Variables for Power-Law Scaling
Statistical Methodology: Errors-in-Variables for Power-Law Scaling
Statistical Methodology: Errors-in-Variables for Power-Law Scaling
Statistical Methodology: Errors-in-Variables for Power-Law Scaling
Statistical Methodology: Errors-in-Variables for Power-Law Scaling
Power-law relationships of the form $Y = \alpha X^{\beta}$ are foundational to ecological theory, but inference on the scaling exponent $\beta$ remains challenging due to two interacting complications: multiplicative measurement error in both variables and habitat-level structural heterogeneity. After logarithmic transformation, the problem reduces to estimating the slope $\beta$ in $y = \alpha_0 + \beta x$, but classical estimators fail to recover the structural exponent when both complications are present.
This paper develops a regime-based benchmarking framework evaluating 20 estimator variants across two data-generating regimes: a single-habitat (homogeneous) regime and a multi-habitat (hierarchical) regime. In the homogeneous setting, orthogonal distance regression achieves near-unbiased recovery, confirming that explicit measurement-error correction is effective when the relationship is common across observations. In the hierarchical setting, however, these same methods exhibit substantial bias and instability (RMSE $\geq 0.503$), while ensemble methods achieve the lowest overall error (RMSE $= 0.1889$). Mixed-effects models, despite explicitly modeling hierarchy, do not recover $\beta$ reliably because they inherit attenuation from conditioning on the error-prone predictor.
These findings demonstrate that inference for power-law scaling cannot be based on measurement-error considerations alone. Estimator choice must be conditioned on the interaction between noise structure and ecological hierarchy. The paper provides a practical framework for method selection and shows that the transition from homogeneity to hierarchy induces a qualitative change in the inferential problem.
This paper develops a regime-based benchmarking framework evaluating 20 estimator variants across two data-generating regimes: a single-habitat (homogeneous) regime and a multi-habitat (hierarchical) regime. In the homogeneous setting, orthogonal distance regression achieves near-unbiased recovery, confirming that explicit measurement-error correction is effective when the relationship is common across observations. In the hierarchical setting, however, these same methods exhibit substantial bias and instability (RMSE $\geq 0.503$), while ensemble methods achieve the lowest overall error (RMSE $= 0.1889$). Mixed-effects models, despite explicitly modeling hierarchy, do not recover $\beta$ reliably because they inherit attenuation from conditioning on the error-prone predictor.
These findings demonstrate that inference for power-law scaling cannot be based on measurement-error considerations alone. Estimator choice must be conditioned on the interaction between noise structure and ecological hierarchy. The paper provides a practical framework for method selection and shows that the transition from homogeneity to hierarchy induces a qualitative change in the inferential problem.
AUTHORS: M. Rajakaruna, D. Talmy, et al.
Submitted to Environmental and Ecological Statistics (Springer Nature)
Theoretical Statistical Physics: Stochastic Rupture and Causal Synchronization
Theoretical Statistical Physics: Stochastic Rupture and Causal Synchronization
Theoretical Statistical Physics: Stochastic Rupture and Causal Synchronization
Theoretical Statistical Physics: Stochastic Rupture and Causal Synchronization
Theoretical Statistical Physics: Stochastic Rupture and Causal Synchronization
Theoretical Statistical Physics: Stochastic Rupture and Causal Synchronization
Collective failure of load-bearing elements appears across scales---from microscopic adhesion clusters to earthquake faults---yet ``simultaneous rupture'' is mathematically ill-defined in stochastic systems. This paper resolves this paradox by introducing a causal observable based on material propagation time $t_{\mathrm{mat}} = L/c_s$. The polarity of mechanical feedback---positive under displacement control, negative under force control---dramatically alters the probability of near-simultaneous failure.
Using a minimal stochastic bond model with tension-dependent hazard $h(T) = r_0 e^{\alpha T}$, we derive exact asymptotic laws for the causal synchronization probability $P_c = \mathbb{P}(\tau < t_{\mathrm{mat}})$. Displacement control yields $P_c \sim (\alpha k v / 2) t_{\mathrm{mat}}$ in the loading-dominated regime, whereas force control with transient unloading gives $P_c \sim r_0 e^{\alpha F} e^{-\alpha F \beta} t_{\mathrm{mat}}$, exponentially suppressing synchronization for any unloading fraction $\beta > 0$. These results identify loading protocol as a control parameter for causal clustering in stochastic rupture, with direct implications for cascade propensity across materials, geophysics, and biological adhesion. A minimal branching-process mapping suggests that displacement control may permit cascading failure ($\mathcal{R} = zp > 1$) while force control stabilizes localized failure ($\mathcal{R} < 1$).
Using a minimal stochastic bond model with tension-dependent hazard $h(T) = r_0 e^{\alpha T}$, we derive exact asymptotic laws for the causal synchronization probability $P_c = \mathbb{P}(\tau < t_{\mathrm{mat}})$. Displacement control yields $P_c \sim (\alpha k v / 2) t_{\mathrm{mat}}$ in the loading-dominated regime, whereas force control with transient unloading gives $P_c \sim r_0 e^{\alpha F} e^{-\alpha F \beta} t_{\mathrm{mat}}$, exponentially suppressing synchronization for any unloading fraction $\beta > 0$. These results identify loading protocol as a control parameter for causal clustering in stochastic rupture, with direct implications for cascade propensity across materials, geophysics, and biological adhesion. A minimal branching-process mapping suggests that displacement control may permit cascading failure ($\mathcal{R} = zp > 1$) while force control stabilizes localized failure ($\mathcal{R} < 1$).
AUTHORS: M. Rajakaruna et al.
Submitted to Physical Review E
Page updated
Google Sites
Report abuse