Discovering genetic biomarkers for targeted cancer therapeutics with eXplainable AI.

Explainable Artificial Intelligence (XAI) enables a holistic understanding of the complex and nonlinear relationships between genes and prognostic outcomes of cancer patients. In this study, we focus on a distinct aspect of XAI - to generate accurate and biologically relevant hypotheses and provide a shorter and more creative path to advance medical research. We present an XAI-driven approach to discover otherwise unknown genetic biomarkers as potential therapeutic targets in high-grade serous ovarian cancer, evidenced by the discovery of IL27RA, which leads to reduced peritoneal metastases when knocked down in tumor-carrying mice given IL27-siRNA-DOPC nanoparticles. Summary: Explainable Artificial Intelligence is amenable to generating biologically relevant testable hypotheses despite their limitations due to explanations originating from post hoc realizations.

Biofilm viscoelasticity and nutrient source location control biofilm growth rate, migration rate, and morphology in shear flow.

We present a numerical model to simulate the growth and deformation of a viscoelastic biofilm in shear flow under different nutrient conditions. The mechanical interaction between the biofilm and the fluid is computed using the Immersed Boundary Method with viscoelastic parameters determined a priori from measurements reported in the literature. Biofilm growth occurs at the biofilm-fluid interface by a stochastic rule that depends on the local nutrient concentration. We compare the growth, migration, and morphology of viscoelastic biofilms with a common relaxation time of 18 min over the range of elastic moduli 10-1000 Pa in different nearby nutrient source configurations. Simulations with shear flow and an upstream or a downstream nutrient source indicate that soft biofilms grow more if nutrients are downstream and stiff biofilms grow more if nutrients are upstream. Also, soft biofilms migrate faster than stiff biofilms toward a downstream nutrient source, and although stiff biofilms migrate toward an upstream nutrient source, soft biofilms do not. Simulations without nutrients show that on the time scale of several hours, soft biofilms develop irregular structures at the biofilm-fluid interface, but stiff biofilms deform little. Our results agree with the biophysical principle that biofilms can adapt to their mechanical and chemical environment by modulating their viscoelastic properties. We also compare the behavior of a purely elastic biofilm to a viscoelastic biofilm with the same elastic modulus of 50 Pa. We find that the elastic biofilm underestimates growth rates and downstream migration rates if the nutrient source is downstream, and it overestimates growth rates and upstream migration rates if the nutrient source is upstream. Future modeling can use our comparison to identify errors that can occur by simulating biofilms as purely elastic structures.

Explainable Artificial Intelligence Reveals Novel Insight into Tumor Microenvironment Conditions Linked with Better Prognosis in Patients with Breast Cancer.

We investigated the data-driven relationship between immune cell composition in the tumor microenvironment (TME) and the ≥5-year survival rates of breast cancer patients using explainable artificial intelligence (XAI) models. We acquired TCGA breast invasive carcinoma data from the cbioPortal and retrieved immune cell composition estimates from bulk RNA sequencing data from TIMER2.0 based on EPIC, CIBERSORT, TIMER, and xCell computational methods. Novel insights derived from our XAI model showed that B cells, CD8+ T cells, M0 macrophages, and NK T cells are the most critical TME features for enhanced prognosis of breast cancer patients. Our XAI model also revealed the inflection points of these critical TME features, above or below which ≥5-year survival rates improve. Subsequently, we ascertained the conditional probabilities of ≥5-year survival under specific conditions inferred from the inflection points. In particular, the XAI models revealed that the B cell fraction (relative to all cells in a sample) exceeding 0.025, M0 macrophage fraction (relative to the total immune cell content) below 0.05, and NK T cell and CD8+ T cell fractions (based on cancer type-specific arbitrary units) above 0.075 and 0.25, respectively, in the TME could enhance the ≥5-year survival in breast cancer patients. The findings could lead to accurate clinical predictions and enhanced immunotherapies, and to the design of innovative strategies to reprogram the breast TME.

Effects of Advective-Diffusive Transport of Multiple Chemoattractants on Motility of Engineered Chemosensory Particles in Fluidic Environments.

Motility behavior of an engineered chemosensory particle (ECP) in fluidic environments is driven by its responses to chemical stimuli. One of the challenges to understanding such behaviors lies in tracking changes in chemical signal gradients of chemoattractants and ECP-fluid dynamics as the fluid is continuously disturbed by ECP motion. To address this challenge, we introduce a new multiscale numerical model to simulate chemotactic swimming of an ECP in confined fluidic environments by accounting for motility-induced disturbances in spatiotemporal chemoattractant distributions. The model accommodates advective-diffusive transport of unmixed chemoattractants, ECP-fluid hydrodynamics at the ECP-fluid interface, and spatiotemporal disturbances in the chemoattractant concentrations due to particle motion. Demonstrative simulations are presented with an ECP, mimicking Escherichia coli (E. coli) chemotaxis, released into initially quiescent fluids with different source configurations of the chemoattractants N-methyl-L-aspartate and L-serine. Simulations demonstrate that initial distributions and temporal evolution of chemoattractants and their release modes (instantaneous vs. continuous, point source vs. distributed) dictate time histories of chemotactic motility of an ECP. Chemotactic motility is shown to be largely determined by spatiotemporal variation in chemoattractant concentration gradients due to transient disturbances imposed by ECP-fluid hydrodynamics, an observation not captured in previous numerical studies that relied on static chemoattractant concentration fields.

Particle Shape Influences Settling and Sorting Behavior in Microfluidic Domains.

We present a new numerical model to simulate settling trajectories of discretized individual or a mixture of particles of different geometrical shapes in a quiescent fluid and their flow trajectories in a flowing fluid. Simulations unveiled diverse particle settling trajectories as a function of their geometrical shape and density. The effects of the surface concavity of a boomerang particle and aspect ratio of a rectangular particle on the periodicity and amplitude of oscillations in their settling trajectories were numerically captured. Use of surrogate circular particles for settling or flowing of a mixture of non-circular particles were shown to miscalculate particle velocities by a factor of 0.9-2.2 and inaccurately determine the particles' trajectories. In a microfluidic chamber with particles of different shapes and sizes, simulations showed that steady vortices do not necessarily always control particle entrapments, nor do larger particles get selectively and consistently entrapped in steady vortices. Strikingly, a change in the shape of large particles from circular to elliptical resulted in stronger entrapments of smaller circular particles, but enhanced outflows of larger particles, which could be an alternative microfluidics-based method for sorting and separation of particles of different sizes and shapes.

Modeling the inactivation of microorganisms occluded in effluent wastewater particles to enhance operation of filtration and disinfection systems.

In disinfection systems, incomplete penetration of chlorine into effluent wastewater particles can result in a residual population of viable microorganisms. In this work, a combined experimental and numerical approach was used to quantify inactivation of microorganisms in effluent particles and identify combinations of particle removal and chlorine dose that would result in a reduction of occluded microorganisms for six full-scale facilities in the United States with different nitrification levels. The results reveal that combined chlorine is more effective for inactivating occluded microorganisms than free chlorine; model calibration results suggest that free chlorine is less effective because it is more reactive. However, nitrified effluents appear to have lower effluent particle concentrations, and decreases in particle concentrations significantly reduce the chlorine required. Additionally, in disinfection systems that are designed and operated based on inactivation of indicator organisms, the chlorine dose may be insufficient to inactivate occluded pathogens to levels consistent with current regulations.

Lattice-Boltzmann simulations of repulsive particle-particle and particle-wall interactions: coughing and choking.

We propose and numerically investigate a new particle retention mechanism for particle entrapment in creeping flows in a constricted section of a saturated rough-walled narrow flow channel. We hypothesize that particles, whose size is smaller than channel width, can be temporarily or permanently immobilized in a flow channel away from channel walls due to particle-particle and particle-wall repulsive potentials, and, consequently, the flow field is clogged temporarily (coughing) or permanently (choking). Two mathematically simplified repulsive particle-particle and particle-wall interaction potentials are incorporated into a two-dimensional colloidal lattice-Boltzmann model. These potentials are two-body Lennard-Jones 12 and screened electrostatic repulsive potentials. Numerical simulations reveal that unlike in smooth-walled flow channels, particles are entrapped away from rough-walled channel walls and subsequently clog the flow field if fluid-drag and repulsive forces on particles are in balance. Off-balance forces, however, could result in temporary clogging if repulsive forces are stronger on the advancing edge of a particle than on its trailing edge. The new conceptualization and two-particle numerical simulations successfully captured (i) temporary entrapment of two particles (coughing), (ii) temporary entrapment of one of the particles with permanent entrapment of the other particle (coughing-choking), and (iii) permanent entrapment of both particles (choking) as a function of repulsive interaction strength.

Effects of etiological agent and bather shedding of pathogens on interpretation of epidemiological data used to establish recreational water quality standards.

The overall goal of the study reported herein was to use techniques in the field of risk assessment (specifically a state-space population dynamic model of disease transmission within recreational waters) to explore the relative significance of (1) active shedding of microorganisms from bathers themselves, and (2) the type and concentration of etiological agent on the observed heterogeneity of the incidence of illness in epidemiological studies that have been used to develop ambient water quality criteria. The etiological agent and corresponding dose ingested during recreational contact was found to significantly impact the observed incidence of illness in an epidemiological study conducted in recreational water. In addition, the observed incidence of illness was found not to necessarily reflect background concentrations of indicator organisms, but rather microorganisms shed during recreational contact. Future revisions to ambient water quality criteria should address the etiological agent, dose, and the significance of microbial shedding relative to background concentrations of pathogens and indicator organisms in addition to the incidence of illness and concentration of indicator organisms. Without a quantitative assessment of these additional variables, study findings may potentially be site specific and not representative of the health risks associated with specific indicator concentrations in all recreational waters.

Robust representation of dry cells in single-layer MODFLOW models.

Dewatered or "dry" grid cells in the USGS ground water modeling software MODFLOW may cause nonphysical artifacts, trigger convergence failures, or interfere with parameter estimation. These difficulties can be avoided in two dimensions by modifying the spatial differencing scheme and the iterative procedure used to resolve nonlinearities. Specifically, the spatial differencing scheme is modified to use the water level on the upstream side of a pair of adjacent cells to calculate the saturated thickness and hence intercell conductance for the pair. This makes it possible to explicitly constrain the water level in a cell to be at or above the cell bottom elevation without introducing nonphysical artifacts. Thus constrained, all initially active cells will remain active throughout the simulation. It was necessary to replace MODFLOW's Picard iteration method with the Newton-Raphson method to achieve convergence in demanding applications involving many dry cells. Tests using a MODFLOW variant based on the new method produced results nearly identical to conventional MODFLOW in situations where conventional MODFLOW converges. The new method is extremely robust and converged in scenarios where conventional MODFLOW failed to converge, such as when almost all cells dewatered. An example application to the Edwards Aquifer in south-central Texas further demonstrates the utility of the new method.

Inactivation of particle-associated microorganisms in wastewater disinfection: modeling of ozone and chlorine reactive diffusive transport in polydispersed suspensions.

Occlusion of microorganisms in wastewater particles often governs the overall performance of a disinfection system, and the associated health risks of post-disinfected effluents. Little is currently known on the penetration of chemical oxidants into particles developed in wastewater treatment. In this work, a reactive transport model that incorporates intra- and extra-particle chemical decay, radial intra-particle diffusion, mass transfer resistance at particle surfaces, and non-linear reaction kinetics within a competitive multi-particle size aqueous environment, was used to analyze the penetration of ozone and chlorine into wastewater particles. Individual characteristics from two secondary wastewater treatment facilities were used in model calibration. Simulations revealed that significant ozone transport within particles greater than 6 microm required large initial concentrations to exhaust the preferential reaction with aqueous soluble matter. Chlorinated samples exhibited apparently slower reactions and thus deeper penetration (22-40 microm). Chlorine penetration was less sensitive to variations in the extra-particle reaction and disinfectant concentration than ozone. Model simulations that considered elevated initial concentrations of chemical disinfectants revealed that complete inactivation of all particle size domains was not possible with current disinfection practices (e.g., contact times). Reduction in the health risks associated with wastewater particles requires treatment that efficiently balances particle removal (filtration) and particle inactivation (disinfection).

Lattice-Boltzmann simulation of coalescence-driven island coarsening.

A two-dimensional lattice-Boltzmann model (LBM) with fluid-fluid interactions was used to simulate first-order phase separation in a thin fluid film. The intermediate asymptotic time dependence of the mean island size, island number concentration, and polydispersity were determined and compared with the predictions of the distribution-kinetics model. The comparison revealed that the combined effects of growth, coalescence, and Ostwald ripening control the phase transition process in the LBM simulations. However, the overall process is dominated by coalescence, which is independent of island mass. As the phase transition advances, the mean island size increases, the number of islands decrease, and the polydispersity approaches unity, which conforms to the predictions of the distribution-kinetics model. The effects of the domain size on the intermediate asymptotic island size distribution, scaling form of the island size distribution, and the crossover to the long-term asymptotic behavior were elucidated.

Preliminary assessment of transport processes influencing the penetration of chlorine into wastewater particles and the subsequent inactivation of particle-associated organisms.

The diffusion of a chemical disinfectant into wastewater particles may be viewed as a serial two-step process involving transport through a macroporous network of pathways to micropores that lead into dense cellular regions. Previous research reveals that ultraviolet (UV) light penetration into wastewater particles is limited primarily to macropores, resulting in a residual concentration of targeted organisms in post-disinfected effluents that reflects the number of organisms embedded in the dense cellular regions of particles. Conversely, chlorine was demonstrated as part of this research to penetrate into both the macroporous and microporous network of pathways, implying that the application of chlorine may be designed feasibly to achieve a desired level of inactivation of particle-associated organisms. In the short term, a disinfection model previously developed for UV irradiation may be used to assess the inactivation of particle-associated organisms with chlorine. However, in the long-term, a more rigorous and complete understanding of the transport of chemical disinfectants into particles can be explored utilizing existing mathematical expressions commonly used to model mass transport into porous media. The parameters of interest in this modeling approach include the reaction rate of chlorine with particulate material, the diffusion rate of chlorine within a particle, the mass-transfer rate coefficient across the particle's boundary, and the particle porosity.

Formulation of a soil-pesticide transport model based on a compartmental approach.

A semianalytical soil-pesticide transport model is formulated based on a compartmental approach to determine spatial and temporal variations of pesticide residues across a soil profile. The compartmental model is implemented by drawing an analogy between a series of continuous-flow stirred tank reactors and a soil horizon that consists of multiple perfectly mixed compartments. The analogy is strengthened by exploiting a relation between the compartment series and the conventional convective-dispersive equation (CDE) for vertical transport in the soil. Consequently, the number of compartments in the model formulation is not free, but dictated as a function of transport parameters. The model formulation allows consideration of arbitrary boundary value specifications and also, for some cases, spatially varying initial concentration profiles. Sorption kinetics is represented via a two-site model that involves a linear sorption isotherm and a first-order irreversible sorption or a radial diffusive penetrating model. For these three cases, analysis of the compartmental model allows the resultant concentration profiles to be expressed in terms of the Poisson distribution. When a nonlinear kinetic sorption model is used to simulate the sorption processes, an analytical solution is not found and a numerical approach is required.