Employee selection research identifies and makes use of associations between individual differences, such as those measured by psychological testing, and individual differences in job performance. Artificial neural networks are computer simulations of biological nerve systems that can be used to model unspecified relationships between sets of numbers. Thirty-five neural networks were trained to estimate normalized annual revenue produced by telephone sales agents based on personality and biographic predictors using concurrent validation data (N=1085). Accuracy of the neural estimates was compared to OLS regression and a proprietary nonlinear model used by the participating company to select agents.

We demonstrate the success of recurrent neural networks in gene network inference and expression prediction using a hybrid of particle swarm optimization and differential evolution to overcome the classic obstacle of local minima in training recurrent neural networks. We also provide an improved validation framework for the evaluation of genetic network modeling systems that will result in better generalization and long-term prediction capability. Success in the modeling of gene regulation and prediction of gene expression will lead to more rapid discovery and development of therapeutic medicine, earlier diagnosis and treatment of adverse conditions, and vast advancements in life science research.

Doctor of Philosophy / Department of Electrical and Computer Engineering / Sanjoy Das / Stochastic algorithms are widely used in various modeling and optimization problems. Evolutionary algorithms are one class of population-based stochastic approaches that are inspired from Darwinian evolutionary theory. A population of candidate solutions is initialized at the first generation of the algorithm. Two variation operators, crossover and mutation, that mimic the real world evolutionary process, are applied on the population to produce new solutions from old ones. Selection based on the concept of survival of the fittest is used to preserve parent solutions for next generation. Examples of such algorithms include genetic algorithm (GA) and genetic programming (GP). Nevertheless, other stochastic algorithms may be inspired from animals’ behavior such as particle swarm optimization (PSO), which imitates the cooperation of a flock of birds. In addition, stochastic algorithms are able to address multi-objective optimization problems by using the concept of dominance. Accordingly, a set of solutions that do not dominate each other will be obtained, instead of just one best solution. This thesis proposes a multi-objective GP-PSO hybrid algorithm to recover gene regulatory network models that take environmental data as stimulus input. The algorithm infers a model based on both phenotypic and gene expression data. The proposed approach is able to simultaneously infer network structures and estimate their associated parameters, instead of doing one or the other iteratively as other algorithms need to. In addition, a non-dominated sorting approach and an adaptive histogram method based on the hypergrid strategy are adopted to address ‘convergence’ and ‘diversity’ issues in multi-objective optimization. Gene network models obtained from the proposed algorithm are compared to a synthetic network, which mimics key features of Arabidopsis flowering control system, visually and numerically. Data predicted by the model are compared to synthetic data, to verify that they are able to closely approximate the available phenotypic and gene expression data. At the end of this thesis, a novel breeding strategy, termed network assisted selection, is proposed as an extension of our hybrid approach and application of obtained models for plant breeding. Breeding simulations based on network assisted selection are compared to one common breeding strategy, marker assisted selection. The results show that NAS is better both in terms of breeding speed and final phenotypic level.

Este estudo tem como objetivo analisar os trade-offs existentes entre os custos logísticos, os incentivos fiscais baseados no ICMS e o custo da neutralização das emissões de carbono geradas nos problemas de localização de Fábricas e Centros de Distribuição. Para isso, elaborou-se um modelo de programação linear inteira mista (PLIM) em GAMS, capaz de determinar o menor custo total de um problema, através da otimização de sua função objetivo composta pelos custos fixos e variáveis dos centros de distribuição e fábricas, custos de transporte (frete de transferência e distribuição), benefícios fiscais e custos ambientais. O modelo foi elaborado de modo a possuir flexibilidade suficiente para simular os diversos cenários que se fizeram necessários durante as análises. Utilizando-se deste modelo, foram avaliados diversos cenários com base em dados reais de uma empresa de bens de consumo não duráveis. Alguns desses cenários estudados mostraram algumas distorções causadas pela existência de incentivos fiscais em alguns Estados brasileiros, mostrando como a guerra fiscal no País pode influenciar decisões estratégicas de negócio. A partir dos resultados obtidos, concluiu-se que o benefício fiscal associado ao crédito presumido de ICMS tem impacto significativo nas decisões de localização, reduzindo de forma relevante os custos totais. Já os custos ambientais, relacionados a neutralização das emissões de carbono, apesar de serem importantes nas decisões de empresas social e ambientalmente responsáveis, possuem peso econômico desprezível e não alteram o resultado da análise. Isso mostra que a política fiscal brasileira gera um aumento da emissão de poluentes na atmosfera e um aumento do desgaste e do fluxo de veículos de transporte pelas rodovias do País. / The main object of this work piece is to analyze existing trade-offs among logistic costs, tax incentives based on ICMS and carbon emission volume variations, to be able to define how these factors influence the network localization of Plants and Distribution Centers. To achieve this objective, a Mixed Integer Linear Programming model was developed in GAMS. The model is able to determine the minimum total cost for a given problem through the optimization of a specific objective function. The components of the objective function are: storage costs, transportation costs (transference and distribution freights), operational fixed costs and tax incentives. The model was designed to have enough flexibility to simulate multiple scenarios required to carry out the analysis. Several logistics configurations were examined using this model. All of the scenarios were established based on real data provided by a consumer goods industry. Nevertheless, some of the studied network configurations are distortions caused by existing tax incentives in some Brazilian states, showing how the fiscal war can influence strategic business decisions. Based on the results, one concludes that the tax benefits associated to the ICMS discounts applied in some Brazilian states actually have significant impact in the location decisions because it cuts down a relevant portion of the operational costs, whereas the carbon credits do not change the chosen network configuration, once it has shown a limited potential for financial benefit. The carbon emissions reduction is, in the other hand, an important aspect of the decisions making in social and environmental responsible companies as it can modify the image of the institution and the way it is perceived by the market.

[pt] O mercado brasileiro de combustíveis apresenta uma nova realidade com a mudança na política de preços praticados por sua principal empresa de petróleo, onde até o ano de 2016 foi caracterizado pelo monopólio devido aos preços artificiais impostos por esta de modo a controlar a inflação no país. Atualmente os preços dos produtos nas refinarias nacionais estão alinhados ao mercado internacional de commodities, viabilizando a entrada de novos competidores para atender a demanda do país. Com este cenário, surgem questões relativas a utilização do refino e níveis de preços a serem adotados no mercado interno de forma a trazer maior competitividade no mercado de forma duradoura e sustentável. Modelos de otimização são utilizados para suportar a tomada de decisão no planejamento da cadeia de downstream e definir a melhor utilização dos recursos disponíveis. Clientes e fornecedores possuem objetivos e custos diferentes, e a necessidade de integrar modelos que dialoguem entre as cadeias de abastecimento destes grupos faz-se necessária, onde os resultados da empresa são impactados de acordo com suas decisões de produção e participação no mercado. O experimento consistiu na construção uma modelagem de rede para a cadeia de distribuição de combustíveis no Brasil a partir de duas ferramentas de otimização existentes, uma delas utilizando-se SIG. Assim, esta modelagem traz uma aplicação eficaz para a empresa, pois a auxilia na quantificação de seus resultados em um cenário de competição em que a mesma se encontra inserida, considerando as singularidades do mercado e indústria no país. / [en] A recent change in the national Brazilian oil company price policy introduced a new market reality as the imposed artificial prices scheme used in order to control inflation was abandoned. Currently, refined products prices in the national territory are matched to the international commodities market, allowing the entry of new competitors to meet national demands. According to this scenario issues relating to infrastructure and a new set of prices to be adopted by the Brazilian domestic market aiming for increased competitiveness on the national market on a long-lasting and sustainable basis begin to appear. Optimization models are used to support the downstream supply chain planning decisionmaking and to ensure the best use of available production resources. Customers and suppliers have different objectives and costs, and it is necessary to integrate models which allow supply chain groups to dialogue among themselves, where the company s result are affected according to their production decisions and market share. The experiment described here consists of the building of a network modelling for the Brazilian fuel distribution chain starting from two optimization tools already available, one of them using GIS. Thus, this modelling brings an effective application to the company, as it assists in the quantification of its results in a competition scenario in which it is inserted, considering the singularities of the Brazilian market and industry.

Le terme symbiose recouvre tous types d'interactions entre espèces et peut être défini comme une association étroite d'espèces différentes vivant ensemble. De telles interactions impliquant des micro-organismes présentent un intérêt particulier pour l'agriculture, la santé, et les questions environnementales. Tous les types d'interactions entre espèces tels que le mutualisme, le commensalisme, et la compétition, sont omniprésents dans la nature et impliquent souvent le métabolisme. La libération de métabolites par des organismes dans l'environnement permet à d'autres individus de la même espèce ou de différentes espèces de les récupérer pour leur usage propre. Dans cette thèse, nous étudions comment les interactions entre espèces façonnentl'environnement. Nous examinons les questions de (i) quels sont les besoins minimaux en éléments nutritifs pour établir la croissance, et (ii) quels métabolites peuvent être échangés entre un organisme et son environnement. L'énumération de tous les ensembles minimaux stoechiométriques de précurseurs et de tous les ensembles minimaux de métabolites échangés,en utilisant des modèles complets de réseaux métaboliques, fournit un meilleur aperçu des interactions entre les espèces. Dans un environnement spatialement homogène, les métabolites qui sont libérés dans un tel environnement sont partagés par tous les individus. Le problème qui se pose alors est de savoir comment les tricheurs, les individus qui profitent des métabolites libérés sans contribuer au bien public, peuvent être exclus de la population. Ceci et d'autres configurations ont déjà été modélisées avec des approches de la théorie des jeux et de l'économie. Nous examinons comment les concepts d'ensembles minimaux de précurseurs stoechiométriques et d'ensembles minimaux de composés échangés peuvent être introduits dans ces modèles / Symbiosis, a term that brings all types of species interaction under one banner, is defined as a close association of different species living together. Species interactions that comprise microorganisms are of particular interest for agriculture, health, and environmental issues. All kinds of species interactions such as mutualism, commensalism, and competition, are omnipresent in nature and occur often at the metabolic level. Organisms release metabolites to the environment which are then taken up by other individuals of the same or of different species. In this thesis, we study how species interactions shape the environment. We examine the questions of (i) what are the minimal nutrient requirements to sustain growth, and (ii) which metabolites can be exchanged between an organism and its environment. Enumerating all minimal stoichiometric precursor sets, and all minimal sets of exchanged metabolites, using metabolic network models, provide a better insight into species interactions. In a spatially homogeneous environment, the metabolites that are released to such an environment are shared by all individuals. The problem that then arises is how cheaters, individuals that profit from the released metabolites without contributing to the public good, can be prevented from the population. This and other configurations were already modeled with approaches from game theory and economics. We examine how the concepts of minimal stoichiometric precursor sets and minimal sets of exchanged compounds can be introduced into such models

Foam reduces gas mobility in porous media by trapping substantial amount of gas and applying a viscous resistance of flowing lamellas to gas flow. In mechanistic foam modeling, gas relative permeability is significantly modified by gas trapping, while an effective gas viscosity, which is a function of flowing lamella density, is assigned to flowing gas. A complete understanding of foam mobility in porous media requires being able to predict the effects of pressure gradient, foam texture, rock and fluid properties on gas trapping, and therefore gas relative permeability, and effective gas viscosity. In the foam literature, separating the contributions of gas trapping and effective gas viscosity on foam mobility has not been achieved because the dynamics of gas trapping and its effects on the effective gas viscosity have been neglected. In this study, dynamics of foam mobility in porous media is investigated with a special focus on gas trapping and its effects on gas relative permeability and effective gas viscosity. Three-dimensional pore-network models representative of real porous media coupled with fluid models characterizing a lamella flow through a pore throat are used to predict flow paths, threshold pressure gradient and Darcy velocity of foam. It is found that the threshold path and the pore volume open above the threshold pressure are independent of the fluid model used in this study. Furthermore, analytical correlations of flowing gas fraction as functions of pressure gradient, lamella density, rock and fluid properties are obtained. At a constant pressure gradient, flowing gas fraction increases as overall lamella density decreases. In the discontinuous-gas foam flow regime, there exists a threshold pressure gradient, which increases with overall lamella density. One of the important findings of this study is that gas relative permeability is a strong non-linear function of flowing gas fraction, opposing most of the existing theoretical models. However, the shape of the relative gas permeability curve is poorly sensitive to overall lamella density. Flowing and trapped lamella densities change with pressure gradient. Moreover, analytical correlations of effective gas viscosity as functions of capillary number, lamella density and rock properties are obtained by up-scaling a commonly used pore-scale apparent gas (lamella) viscosity model. Effective gas viscosity increases nonlinearly with flowing lamella density, which opposes to the existing linear foam viscosity models. In addition, the individual contributions of gas trapping and effective gas viscosity on foam mobility are quantified for the first time. The functional relationship between effective gas viscosity and flowing lamella density in the presence of dynamic trapped gas is verified. A mechanistic foam model is developed by using the analytical correlations of flowing gas fraction and effective gas viscosity generated from the pore-network study and a modified population balance model. The developed model is successful in simulating unsteady-state and steady state flow of foam through porous media. Moreover, the flow behaviors in high- and low-quality flow regimes are verified by the experimental studies in the literature. Finally, the simulation results are successfully history matched with two different core-flood data. / text

Genetic regulatory networks can model dynamics of cells. They also allow for studying the effect of internal or external interventions. Selectively applying interventions towards a certain objective is known as controlling network dynamics. In this thesis work, the issue of how the external interventions af fect the network is studied. The effects are determined using differential gene expression analysis. The differential gene expression problem is further studied to improve the power of the given method. Control problem for dynamic discrete regulatory networks is formulated. This also addresses the needs for various control strategies, e.g., finite horizon, infinite horizon, and various accounting of state and intervention costs. Control schemes for small to large networks are proposed and experimented. A case study is provided to show how the proposals are exploited / also given is the need for and effectiveness of various control schemes.

Este estudo tem como objetivo analisar os trade-offs existentes entre os custos logísticos, os incentivos fiscais baseados no ICMS e o custo da neutralização das emissões de carbono geradas nos problemas de localização de Fábricas e Centros de Distribuição. Para isso, elaborou-se um modelo de programação linear inteira mista (PLIM) em GAMS, capaz de determinar o menor custo total de um problema, através da otimização de sua função objetivo composta pelos custos fixos e variáveis dos centros de distribuição e fábricas, custos de transporte (frete de transferência e distribuição), benefícios fiscais e custos ambientais. O modelo foi elaborado de modo a possuir flexibilidade suficiente para simular os diversos cenários que se fizeram necessários durante as análises. Utilizando-se deste modelo, foram avaliados diversos cenários com base em dados reais de uma empresa de bens de consumo não duráveis. Alguns desses cenários estudados mostraram algumas distorções causadas pela existência de incentivos fiscais em alguns Estados brasileiros, mostrando como a guerra fiscal no País pode influenciar decisões estratégicas de negócio. A partir dos resultados obtidos, concluiu-se que o benefício fiscal associado ao crédito presumido de ICMS tem impacto significativo nas decisões de localização, reduzindo de forma relevante os custos totais. Já os custos ambientais, relacionados a neutralização das emissões de carbono, apesar de serem importantes nas decisões de empresas social e ambientalmente responsáveis, possuem peso econômico desprezível e não alteram o resultado da análise. Isso mostra que a política fiscal brasileira gera um aumento da emissão de poluentes na atmosfera e um aumento do desgaste e do fluxo de veículos de transporte pelas rodovias do País. / The main object of this work piece is to analyze existing trade-offs among logistic costs, tax incentives based on ICMS and carbon emission volume variations, to be able to define how these factors influence the network localization of Plants and Distribution Centers. To achieve this objective, a Mixed Integer Linear Programming model was developed in GAMS. The model is able to determine the minimum total cost for a given problem through the optimization of a specific objective function. The components of the objective function are: storage costs, transportation costs (transference and distribution freights), operational fixed costs and tax incentives. The model was designed to have enough flexibility to simulate multiple scenarios required to carry out the analysis. Several logistics configurations were examined using this model. All of the scenarios were established based on real data provided by a consumer goods industry. Nevertheless, some of the studied network configurations are distortions caused by existing tax incentives in some Brazilian states, showing how the fiscal war can influence strategic business decisions. Based on the results, one concludes that the tax benefits associated to the ICMS discounts applied in some Brazilian states actually have significant impact in the location decisions because it cuts down a relevant portion of the operational costs, whereas the carbon credits do not change the chosen network configuration, once it has shown a limited potential for financial benefit. The carbon emissions reduction is, in the other hand, an important aspect of the decisions making in social and environmental responsible companies as it can modify the image of the institution and the way it is perceived by the market.

Accurate prediction of colloid deposition rates in porous media is essential in several applications. These include natural filtration of pathogenic microorganisms such as bacteria, viruses, and protozoa, transport and fate of colloid-associated transport of contaminants, deep bed and river bank filtration for water treatment, fate and transport of engineered nanoparticles released into the environment, and bioremediation of contaminated sites. Colloid transport in porous media is a multi-scale problem, with length scales spanning from the sub-pore scale, where the particle-soil interaction forces control the deposition, up to the Darcy scale, where the macroscopic equations governing particle transport are formulated. Colloid retention at the Darcy scale is due to the lumped effect of processes occurring at the pore scale. This requires the incorporation of the micro-scale physics into macroscopic models for a better understanding of colloid deposition in porous media. That can be achieved through pore-scale modeling and the subsequent upscaling to the Darcy scale. Colloid Filtration Theory (CFT), the most commonly used approach to describe colloid attachment onto the soil grains in the subsurface, is found to accurately predict the deposition rates of micron-sized particles under favorable conditions for deposition. But, CFT has been found to over predict particle deposition rates at low flow velocity conditions, typical of groundwater flow, and for nanoscale particles. Also, CFT is found to be inapplicable at typical environmental conditions, where conditions become unfavorable for deposition, due to factors not considered in CFT such as deposition in the secondary minimum of the interaction energy profile, grain surface roughness, surface charge heterogeneity of grains and colloids, and deposition at grain-to-grain contacts. To the best of our knowledge, mechanistic-based models for predicting colloid deposition rates under unfavorable conditions do not exist. Currently, fitting the colloid breakthrough curve (BTC), obtained from the laboratory column-or field-scale experiments, to the advection-dispersion-deposition model is used to estimate the values of deposition rate coefficients. Because of their small size (less than 100 nm), nanoparticles, a sub-class of colloids, may interact with the porous medium in a different way as compared to the larger colloids, resulting in different retention mechanisms for nanoparticles and micron-sized particles. This emphasizes the need to study nanoparticles separately from larger, micrometer-sized colloids to better understand nanoparticle retention mechanisms. The work reported in this thesis contributes towards developing mathematical models to predict nanoparticle movement in porous media. A comprehensive mechanistic approach is employed by integrating pore-scale processes into Darcy-scale models through pore-network modeling to upscale nanoparticle transport in saturated porous media to the Darcy scale, and to develop correlation equations for the Darcy-scale deposition parameters in terms of various measurable parameters at Darcy scale. Further, a one-dimensional mathematical model to simulate the co-transport of viruses and colloids in partially saturated porous media is developed to understand the relative importance of various interactions on virus transport in porous media. Pore-network modeling offers a valuable upscaling tool to express the macroscopic behavior by accounting for the relevant physics at the underlying pore scale. This is done by idealizing the pore space as an interconnected network of pore elements of different sizes and variably connected to each other, and simulating flow and transport through the network of pores, with the relevant physics implemented on a pore to pore basis (Raoof, 2011). By comparing the results of pore-network modeling with an appropriate mathematical model describing the macro-scale behavior, a relationship between the properties at the macro scale and those at the pore scale can be obtained. A three dimensional multi-directional pore-network model, PoreFlow, developed by Raoof et al. (2010, 2013) is employed in this thesis, which represents the porous medium as an interconnected network of cylindrical pore throats and spherical pore bodies, to upscale nanoparticle transport from pore scale to the Darcy scale. The first step in this procedure is to obtain relationships between adsorbed mass and aqueous mass for a single pore. A mathematical model is developed to simulate nanoparticle transport in a saturated cylindrical pore by solving the full transport equation, considering various processes such as advection, diffusion, hydrodynamic wall effects, and nanoparticle-collector surface interactions. The pore space is divided into three different regions: bulk, diffusion and potential regions, based on the dominant processes acting in each of these regions. In both bulk and diffusion regions, nanoparticle transport is governed by advection and diffusion. However, in the diffusion region, the diffusion is significantly reduced due to hydrodynamic wall effects. Nanoparticle-collector interaction forces dominate the transport in the potential region where deposition occurs. A sensitivity analysis of the model indicates that nanoparticle transport and deposition in a pore is significantly affected by various pore-scale parameters such as the nanoparticle and collector surface potentials, ionic strength of the solution, flow velocity, pore radius, and nanoparticle radius. The model is found to be more sensitive to all parameters under favorable conditions. It is found that the secondary minimum plays an important role in the deposition of small as well as large nanoparticles, and its contribution is found to increase as the favorability of the surface for adsorption decreases. Correlation equations for average deposition rate coefficients of nanoparticles in a saturated cylindrical pore under unfavorable conditions are developed as a function of nine pore-scale parameters: the pore radius, nanoparticle radius, mean flow velocity, solution ionic strength, viscosity, temperature, solution dielectric constant, and nanoparticle and collector surface potentials. Advection-diffusion equations for nanoparticle transport are prescribed for the bulk and diffusion regions, while the interaction between the diffusion and potential regions is included as a boundary condition. This interaction is modeled as a first-order reversible kinetic adsorption. The expressions for the mass transfer rate coefficients between the diffusion and the potential regions are derived in terms of the interaction energy profile between the nanoparticle and the collector. The resulting equations are solved numerically for a range of values of pore-scale parameters. The nanoparticle concentration profile obtained for the cylindrical pore is averaged over a moving averaging volume within the pore in order to get the 1-D concentration field. The latter is fitted to the 1-D advection-dispersion equation with an equilibrium or kinetic adsorption model to determine the values of the average deposition rate coefficients. Pore-scale simulations are performed for three values of Péclet number, Pe = 0.05, 5 and 50. It is found that under unfavorable conditions, the nanoparticle deposition at pore scale is best described by an equilibrium model at low Péclet numbers (Pe = 0.05), and by a kinetic model at high Péclet numbers (Pe = 50). But, at an intermediate Pe (e.g., near Pe = 5), both equilibrium and kinetic models fit the 1-D concentration field. Correlation equations for the pore-averaged nanoparticle deposition rate coefficients under unfavorable conditions are derived by performing a multiple-linear regression analysis between the estimated deposition rate coefficients for a single pore and various pore-scale parameters. The correlation equations, which follow a power law relationship with nine pore-scale parameters, are found to be consistent with the column-scale and pore-scale experimental results, and qualitatively agree with CFT. Nanoparticle transport is upscaled from pore to the Darcy scale in saturated porous media by incorporating the correlations equations for the pore-averaged deposition rate coefficients of nanoparticles in a cylindrical pore into a multi-directional pore-network model, PoreFlow (Raoof et al., 2013). Pore-network model simulations are performed for a range of parameter values, and nanoparticle BTCs are obtained from the pore-network model. Those curves are then modeled using 1-D advection-dispersion equation with a two-site first-order reversible deposition, with terms accounting for both equilibrium and kinetic sorption. Kinetic sorption is found to become important as the favorability of the surface for deposition decreases. Correlation equations for the Darcy¬scale deposition rate coefficients under unfavorable conditions are developed as a function of various measurable Darcy-scale parameters, including: porosity, mean pore throat radius, mean pore water velocity, nanoparticle radius, ionic strength, dielectric constant, viscosity, temperature, and surface potentials on the nanoparticle and grain surface. The correlation equations are found to be consistent with the observed trends from the column experiments available in the literature, and are in agreement with CFT for all parameters, except for the mean pore water velocity and nanoparticle radius. The Darcy-scale correlation equations contain multipliers whose values for a given set of experimental conditions need to be determined by comparing the values of the deposition rate coefficients predicted by the correlation equations against the estimated values of Darcy-scale deposition parameters obtained by fitting the BTCs from column or field experiments with 1-D advection-dispersion-deposition model. They account for the effect of factors not considered in this study, such as the physical and chemical heterogeneity of the grain surface and nanoparticles, flow stagnation points, grain-to-grain contacts, etc. Colloids are abundant in the subsurface and have been observed to interact with a variety of contaminants, including viruses, thereby significantly influencing their transport. A mathematical model is developed to simulate the co-transport of viruses and colloids in partially saturated porous media under steady state flow conditions. The virus attachment to the mobile and immobile colloids is described using a linear reversible kinetic model. It is assumed that colloid transport is not affected by the presence of attached viruses on its surface, and hence, colloid transport is decoupled from virus transport. The governing equations are solved numerically using an alternating three-step operator splitting approach. The model is verified by fitting three sets of experimental data published in the literature: (1) Syngouna and Chrysikopoulos (2013) and (2) Walshe et al. (2010), both on the co-transport of viruses and clay colloids under saturated conditions, and (3) Syngouna and Chrysikopoulos (2015) for the co-transport of viruses and clay colloids under unsaturated conditions. The model results are found to be in good agreement with the observed BTCs under both saturated and unsaturated conditions. Then, the developed model was used to simulate the co-transport of viruses and colloids in porous media under unsaturated conditions, with the aim of understanding the relative importance of various processes on the co-transport of viruses and colloids. The virus retention in porous media in the presence of colloids is greater under unsaturated conditions as compared to the saturated conditions due to: (1) virus attachment to the air-water interface (AWI), and (2) co-deposition of colloids with attached viruses on its surface to the AWI. A sensitivity analysis of the model to various parameters showed that virus attachment to AWI is the most sensitive parameter affecting the BTCs of both free viruses and total mobile viruses, and has a significant effect on all parts of the BTC. The free and the total mobile virus BTCs are mainly influenced by parameters describing virus attachment to the AWI, virus interactions with mobile and immobile colloids, virus attachment to solid-water interface (SWI), and colloid interactions with SWI and AWI. The virus BTC is relatively insensitive to parameters describing the maximum adsorption capacity of the AWI for colloids, inlet colloid concentration, virus detachment rate coefficient from the SWI, maximum adsorption capacity of the AWI for viruses, and inlet virus concentration.