Bio papers — 2026-10-15

Today we are diving into modeling the microenvironmental dynamics of adenoid cystic carcinoma in the lacrimal gland because this rare and life-threatening tumor has very limited treatment options, making understanding its biology crucial for better care. The work by AKMPC develops a mathematical model using nonlinear ordinary differential equations to show that tumor cell growth is linked to oxygen and nutrient depletion. Severe hypoxia drives abnormal blood vessel formation, which complicates control. This finding suggests that while surgery helps, postoperative radiotherapy offers better control if the residual tumor cells are radiosensitive and treatment interruptions are avoided.

Moving down in importance, there is research into simulating Alzheimer's amyloid-like pathology on a fly central nervous system connectome. This simulation showed that the amount of lost neurons does not accurately predict functional loss. Instead, whether damage hits poorly connected or highly connected neurons determines the deficit in whole-brain activity. This contrasts with how many studies view pathology as a simple measure of neuron loss.

Another piece of work explores neural computation by looking at how observational resolution shapes the apparent structure of neural computation. This suggests that projection can convert higher-order dependence into pairwise structure. This idea is related to how we interpret information in complex systems.

Finally, there is modeling environmental DNA dynamics in river environments using a stochastic partial differential equation model to track fish abundance based on migration dynamics. This framework provides a first step for understanding eDNA concentration, though the mechanistic uncertainties remain a key open question.

The work by State Space Models is particularly important because it suggests a mathematical way to generate complex temporal patterns like time cells and oscillations directly from structure. This mirrors how abstract cognitive functions might emerge in the brain. This framework uses a structured multi-timescale complex rotational basis to naturally create these dynamics through optimized history compression. It can scale up to model things like event counting, providing a computationally tractable lens for understanding neural activity.

Building on this idea, the approach presented in Tracking large chemical reaction networks and rare events by neural networks tackles the intractability of modeling high-dimensional stochastic dynamics using chemical master equations. This method uses an advanced neural network that learns from autoregressive representations to efficiently train on complex systems, such as a genetic toggle switch with sixteen species and thirty-five reactions. Furthermore, this technique allows for the quantification of rare events in spatially extended models up to system sizes reaching three times ten to the power of fifteen.

In terms of biological context, imaging-anchored Multiomics in Cardiovascular Disease offers a way to link clinical imaging phenotypes with molecular states by integrating cardiac MRI and transcriptomics. This work reviews representation learning strategies for multimodal fusion, focusing on how to handle missing data and batch effects when combining these different data types. This connects to the idea of capturing context. Similarly, Deep learning-based virtual perturbation reveals increasing heterogeneity of gene regulatory responses with aging by predicting cell-specific effects from unperturbed transcriptomes.

Finally, the work on Synergy and Complementarity: The Generative Basis of Chemical Organizations addresses how reaction networks maintain persistence by defining minimal building blocks called elementary reaction closures. This study shows that relevant persistent modules can be built from these minimal sequences. It quantifies synergy and complementarity as criteria for generating these structures across hundreds of biological networks.

The most critical issue right now is how we are actually testing surgical AI tools because the current validation practices are fundamentally flawed, leading to potentially dangerous clinical adoption issues. This work, a catalogue of validation pitfalls derived from expert consensus, shows that existing methods often fail to respect the temporal and hierarchical nature of intraoperative videos. This framework provides an evidence-based structure for how we should benchmark these algorithms and prepare them for regulatory review and actual use in a clinical setting. It highlights widespread problems across data handling, metric selection, and how results are reported.

One key finding from this work is that ignoring the temporal and hierarchical structures in surgical video datasets can significantly underestimate uncertainty. It can also hide important failure modes and even change how algorithms are ranked. This is a major problem because it means we might trust a tool more than we should based on incomplete testing.

Another piece of work delves into modeling anomalous diffusion using fractional Brownian motion, specifically estimating the Hurst exponent H and diffusion coefficient D H from density measurements and particle trajectories. This research shows that inferring both parameters from just one measurement in time is impossible, emphasizing the need for multiple time points for good precision.

Furthermore, a neural network trained with a squared error loss can approximate the posterior mean of H very cheaply. This finding is useful because it can be computed almost without cost. It connects to how we might try to model complex motion patterns, as seen in studies on vesicle trajectories where H and D H form a continuum rather than fixed modes.

Today's papers

Modeling the Microenvironmental Dynamics of Adenoid Cystic Carcinoma in Lacrimal Gland with Treatment Intervention A mathematical model is developed to understand how tumor growth relates to oxygen depletion and blood vessel formation, suggesting postoperative radiotherapy combined with good radiosensitivity can reduce recurrence risk. An Open-Source Web-Based Viewer for Disseminating Image-Derived Models Developed a web viewer that allows clinicians to access patient-specific 3D models from any device without needing local installation or specialized hardware. Stochastic partial differential equation model for environmental DNA dynamics in river environments A new mathematical framework uses a stochastic partial differential equation to model how environmental DNA concentration changes in rivers based on fish migration dynamics. EEG Correlates of Verbal and Perceptual Reasoning This study used quantitative EEG recordings to check if brain wave patterns correlate with performance on verbal and perceptual reasoning tests but found no significant relationships. How Long Is the Coast of a Neuron? Resolution-Dependent Interaction Order and Computational Capacity This paper suggests that how we measure neural computations changes the apparent structure of the computation depending on the resolution used. Effects of cortisol time-scales in a model of emotion regulation This study shows that how quickly cortisol acts on brain circuits matters as much as what it does, depending on its temporal memory. Reciprocity coordinates a three-way trade-off in recurrent neural circuits This research found that increasing the reciprocity between neurons creates a trade-off between their capacity to store information, their stability, and how well they can separate different task states. Simulating Alzheimer's amyloid-like pathology on a fly central nervous system connectome This study used a simulated fly brain model to show that the way neurons are damaged matters more than just the number of neurons lost when studying disease effects. State Space Models Naturally Produce Time Cell and Oscillatory Behaviors and Scale to Abstract Cognitive Functions This paper proposes using state space models to generate complex time-based behaviors like time cells by leveraging a structured multi-timescale basis for neural activity. Tracking large chemical reaction networks and rare events by neural networks This method uses a neural network approach to efficiently simulate high-dimensional chemical reaction networks and rare events, even in very large systems. Imaging-anchored Multiomics in Cardiovascular Disease: Integrating Cardiac Imaging, Bulk, Single-cell, and Spatial Transcriptomics This review discusses how to combine cardiac imaging with molecular data from different cell types and locations to better understand cardiovascular disease. How deterministic trajectories and their fluctuations underlie allele-frequency statistics in the strong-selection regime This work develops a framework to calculate allele frequency statistics by conditioning on whether an allele eventually fixes or is lost under strong selection. Cancer genomics, targeted therapy to precision oncology: the treatment paradigm is a holistic approach to cancer therapeutics This article explains how precision oncology uses genetic profiling to tailor cancer treatments by targeting specific molecular alterations. Deep learning-based virtual perturbation reveals increasing heterogeneity of gene regulatory responses with aging This study uses deep learning to show that aging causes gene regulatory responses to become more varied and less consistent between cells. Physiology-Informed Sparse Identification of Dynamic Crop Models This work proposes using sparse identification methods to find simple mathematical models that describe how crops grow in response to environmental changes and different plant types. Synergy and Complementarity: The Generative Basis of Chemical Organizations This paper explores how small, fundamental building blocks combine in specific ways to create complex, persistent structures in biological reaction networks. Current validation practice undermines surgical AI development This work provides a guide on how to rigorously validate surgical AI by identifying common pitfalls related to data, metrics, and reporting in surgical videos. Maximum likelihood, neural networks and posterior mean for inferring fractional Brownian motion parameters This study uses neural networks to approximate the posterior mean of parameters describing anomalous diffusion from density measurements and trajectories. An open-source web-based viewer for disseminating image-derived models Developed a web viewer that allows clinicians to access patient-specific 3D models from any device without needing local installation or specialized hardware. [paper]
The paper's title is Modeling the Microenvironmental Dynamics of Adenoid Cystic Carcinoma in Lacrimal Gland with Treatment Intervention A mathematical model is developed to understand how tumor growth relates to oxygen depletion and blood vessel formation, suggesting postoperative radiotherapy combined with good radiosensitivity can reduce recurrence risk.
The paper's title is An Open-Source Web-Based Viewer for Disseminating Image-Derived Models Developed a web viewer that allows clinicians to access patient-specific 3D models from any device without needing local installation or specialized hardware.
The paper's title is Stochastic partial differential equation model for environmental DNA dynamics in river environments A new mathematical framework uses a stochastic partial differential equation to model how environmental DNA concentration changes in rivers based on fish migration dynamics.
The paper's title is EEG Correlates of Verbal and Perceptual Reasoning This study used quantitative EEG recordings to check if brain wave patterns correlate with performance on verbal and perceptual reasoning tests but found no significant relationships.
The paper's title is How Long Is the Coast of a Neuron? Resolution-Dependent Interaction Order and Computational Capacity This paper suggests that how we measure neural computations changes the apparent structure of the computation depending on the resolution used.
The paper's title is Effects of cortisol time-scales in a model of emotion regulation This study shows that how quickly cortisol acts on brain circuits matters as much as what it does, depending on its temporal memory.
The paper's title is Reciprocity coordinates a three-way trade-off in recurrent neural circuits This research found that increasing the reciprocity between neurons creates a trade-off between their capacity to store information, their stability, and how well they can separate different task states.
The paper's title is Simulating Alzheimer's amyloid-like pathology on a fly central nervous system connectome This study used a simulated fly brain model to show that the way neurons are damaged matters more than just the number of neurons lost when studying disease effects.
The paper's title is State Space Models Naturally Produce Time Cell and Oscillatory Behaviors and Scale to Abstract Cognitive Functions This paper proposes using state space models to generate complex time-based behaviors like time cells by leveraging a structured multi-timescale basis for neural activity.
The paper's title is Tracking large chemical reaction networks and rare events by neural networks This method uses a neural network approach to efficiently simulate high-dimensional chemical reaction networks and rare events, even in very large systems.
The paper's title is Imaging-anchored Multiomics in Cardiovascular Disease: Integrating Cardiac Imaging, Bulk, Single-cell, and Spatial Transcriptomics This review discusses how to combine cardiac imaging with molecular data from different cell types and locations to better understand cardiovascular disease.
The paper's title is How deterministic trajectories and their fluctuations underlie allele-frequency statistics in the strong-selection regime This work develops a framework to calculate allele frequency statistics by conditioning on whether an allele eventually fixes or is lost under strong selection.
The paper's title is Cancer genomics, targeted therapy to precision oncology: the treatment paradigm is a holistic approach to cancer therapeutics This article explains how precision oncology uses genetic profiling to tailor cancer treatments by targeting specific molecular alterations.
The paper's title is Deep learning-based virtual perturbation reveals increasing heterogeneity of gene regulatory responses with aging This study uses deep learning to show that aging causes gene regulatory responses to become more varied and less consistent between cells.
The paper's title is Physiology-Informed Sparse Identification of Dynamic Crop Models This work proposes using sparse identification methods to find simple mathematical models that describe how crops grow in response to environmental changes and different plant types.
The paper's title is Synergy and Complementarity: The Generative Basis of Chemical Organizations This paper explores how small, fundamental building blocks combine in specific ways to create complex, persistent structures in biological reaction networks.
The paper's title is Current validation practice undermines surgical AI development This work provides a guide on how to rigorously validate surgical AI by identifying common pitfalls related to data, metrics, and reporting in surgical videos.

The papers

Important terms

Hypoxia: This refers to a lack of oxygen in a tissue, which is shown to drive abnormal blood vessel formation and tumor cell growth in adenoid cystic carcinoma.
Central Nervous System Connectome: This involves simulating the connections within the brain's nervous system, showing that how damage hits neurons matters more than just counting lost neurons.
Stochastic Partial Differential Equation Model: A mathematical model used to track environmental DNA dynamics in rivers, helping to understand fish abundance based on their movement and concentration.
Neural Computation Resolution: This concept explores how the way we observe neural activity affects what we perceive as its structure, suggesting projection can simplify complex dependencies.