arXiv subjects

Daniel Coombs

Publications and source records attributed to Daniel Coombs.

3 recordsLinked to original sources

arXiv2026

Mean First Passage Time for Persistent Random Walks in Annular Search Domains

We study the mean first-passage time of a random walker to a small absorbing target at the center of a two-dimensional annulus with a specularly reflecting outer boundary. The problem is motivated by natural killer cell migration toward a target cancer cell, where the goal is to quantify how long it takes immune cells to reach the target and how search efficiency depends on directional persistence and chemotactic bias. Cell motion is modeled as a velocity-jump process. We first consider a correlated random walk with a von Mises turning kernel, with a concentration parameter controlling directional persistence. We then extend the model to a biased correlated random walk using a phase-shifted turning kernel that represents preferential motion, for example following a concentration gradient. Our analysis combines closed-form benchmarks for simple and biased random walks, Fourier-mode reductions of the transport equations for the correlated and biased correlated models, and a fast-turning perturbation expansion that gives an analytical correction to the diffusion-limit mean first-passage time for the random walker. Our analytical results are supported by numerical methods that include a semi-Lagrangian solver in radial and angular coordinates, a stationary discretisation designed to handle biased transport, and an event-driven Monte Carlo simulator for cross-validation. Together, our results provide a framework relating persistent and biased immune-cell motion to target-search times in confined two-dimensional domains.

q-bio.CB↗

arXiv2025

Reward Engineering for Spatial Epidemic Simulations: A Reinforcement Learning Platform for Individual Behavioral Learning

We present ContagionRL, a Gymnasium-compatible reinforcement learning platform specifically designed for systematic reward engineering in spatial epidemic simulations. Unlike traditional agent-based models that rely on fixed behavioral rules, our platform enables rigorous evaluation of how reward function design affects learned survival strategies across diverse epidemic scenarios. ContagionRL integrates a spatial SIRS+D epidemiological model with configurable environmental parameters, allowing researchers to stress-test reward functions under varying conditions including limited observability, different movement patterns, and heterogeneous population dynamics. We evaluate five distinct reward designs, ranging from sparse survival bonuses to a novel potential field approach, across multiple RL algorithms (PPO, SAC, A2C). Through systematic ablation studies, we identify that directional guidance and explicit adherence incentives are critical components for robust policy learning. Our comprehensive evaluation across varying infection rates, grid sizes, visibility constraints, and movement patterns reveals that reward function choice dramatically impacts agent behavior and survival outcomes. Agents trained with our potential field reward consistently achieve superior performance, learning maximal adherence to non-pharmaceutical interventions while developing sophisticated spatial avoidance strategies. The platform's modular design enables systematic exploration of reward-behavior relationships, addressing a knowledge gap in models of this type where reward engineering has received limited attention. ContagionRL is an effective platform for studying adaptive behavioral responses in epidemic contexts and highlight the importance of reward design, information structure, and environmental predictability in learning. Our code is publicly available at https://github.com/redradman/ContagionRL

cs.LG↗

arXiv2010

A biophysical model of cell adhesion mediated by immunoadhesin drugs and antibodies

A promising direction in drug development is to exploit the ability of natural killer cells to kill antibody-labeled target cells. Monoclonal antibodies and drugs designed to elicit this effect typically bind cell-surface epitopes that are overexpressed on target cells but also present on other cells. Thus it is important to understand adhesion of cells by antibodies and similar molecules. We present an equilibrium model of such adhesion, incorporating heterogeneity in target cell epitope density and epitope immobility. We compare with experiments on the adhesion of Jurkat T cells to bilayers containing the relevant natural killer cell receptor, with adhesion mediated by the drug alefacept. We show that a model in which all target cell epitopes are mobile and available is inconsistent with the data, suggesting that more complex mechanisms are at work. We hypothesize that the immobile epitope fraction may change with cell adhesion, and we find that such a model is more consistent with the data. We also quantitatively describe the parameter space in which binding occurs. Our results point toward mechanisms relating epitope immobility to cell adhesion and offer insight into the activity of an important class of drugs.

q-bio.CB↗