Bio papers — 2026-10-14

Today's work centers on ensuring thermodynamic consistency in parameter optimization for biochemical reaction networks, which is crucial because it provides a rigorous framework for understanding how these systems behave under different conditions. This is addressed by the approach described in AKMPC, which tackles enforcing thermodynamic consistency during parameter optimization.

A key piece of research involves quantifying uncertainty in artificial intelligence agreement regarding breast-cancer pathology using reference labels that depend on the reader and annotation conditions, as detailed in Bayesian Uncertainty Quantification of Breast-Cancer Pathology AI Agreement under Reader- and Annotation-Condition-Dependent Reference Labels. This work helps us understand the reliability of AI outputs when the input labels themselves are context dependent.

Moving down, there is a computational framework called EPITIME, which uses structure preserving discretizations for integral epidemic models to handle complex dynamics, as presented in EPITIME: A Computational Framework for Integral Epidemic Models with Structure-Preserving Discretizations. This is important because it allows us to model disease spread more accurately by preserving the underlying structure of the system during discretization.

Another area explored is overcoming the curse of dimensionality when reconstructing structural connectivity in nonlinear networks through pairwise information flow, as seen in Overcoming the Curse of Dimensionality: Structural Connectivity Reconstruction via Pairwise Information Flow in Nonlinear Networks. This helps us map out how different parts of a complex network are connected without getting lost in too much data.

Finally, there is work on predicting clone-specific perturbation responses using lineage supervised neural inverse optimal transport, which is CloneOT CloneOT: predicting clone-specific perturbation responses with lineage-supervised neural inverse optimal transport. This method predicts how specific genetic changes will affect a particular lineage, building on the idea of understanding functional element composition across different organisms mentioned in Homogeneity and heterogeneity of functional element composition of genome units across human and seven model organisms.

The most pressing work today seems to be the development of RNAGenScape, which attempts to guide the generation of mRNA sequences using property-guided manifold Langevin dynamics, a method that matters because it promises better control over synthetic biology outputs. This approach builds upon earlier work like Discovery of Bioresorbable Polymer Suture Coatings for Controlled Tissue Regeneration with Multimodal Foundation Models, which explored using multimodal foundation models to discover bioresorbable polymer suture coatings for controlled tissue regeneration.

A significant piece of progress involves uncertainty-aware integration of deuterium metabolic imaging with genome-scale metabolic modeling, which reveals how diet influences the feasible space remodeling in mouse skeletal muscle Uncertainty-aware integration of deuterium metabolic imaging with genome-scale metabolic modeling reveals diet-associated feasible-space remodeling in mouse skeletal muscle. This work is important because it connects dietary inputs to physiological outcomes at the molecular level, and it relates back to how other models handle complex systems.

Then there is the ensemble-based modelling of optogenetic bioprocesses under phototoxicity Ensemble-Based Modelling of Optogenetic Bioprocesses under Phototoxicity, which tackles the challenges of modeling biological processes when dealing with light-induced damage. This is a necessary step before we can accurately predict outcomes in areas like synthetic gene expression.

We also see some foundational work on the regularity of reaction systems On the Regularity of Reaction Systems, which provides mathematical structure for understanding how these biological reactions behave. This underlying structure is crucial for making the Langevin dynamics used in RNAGenScape more robust.

Finally, there is work on disease-level drug approval forecasting from trial-agnostic public data Disease-level drug approval forecasting from trial-agnostic public data, which attempts to predict drug approvals without relying on specific clinical trials. This speaks to the broader goal of using large datasets for predictive modeling across different domains.

The most compelling work today is the mathematical model for fitness effects on viral persistence because it directly addresses how viruses maintain their presence, which is a huge question in infectious disease dynamics. This model explores how viral persistence changes based on different factors, and it connects to the ideas in A mathematical model for fitness effects on viral persistence by quantifying these fitness consequences.

We also saw some interesting work on consensus-level substitution rates being distinct from the virion-level rate, which provides a finer resolution to how genetic changes accumulate across different scales Consensus-level substitution rates are distinct from the virion-level rate. This distinction is important because it suggests that simple measures of mutation might not capture the full picture of evolutionary dynamics, and this finding feeds into how we understand long-term population stability.

Then there is the robust numerical framework for multi-strain epidemics, which offers a solid computational tool for simulating complex disease spread A Robust Numerical Framework for Multi-Strain Epidemics. This framework helps us test scenarios involving multiple strains simultaneously, and it builds upon the foundational modeling concepts discussed in The influence of decay in dynamic models of gene regulatory networks by incorporating temporal dynamics into the network structure.

Finally, we have some work on effective transport statistics for cooperative molecular motor ensembles with tether slack, which sheds light on how physical constraints affect movement Effective transport statistics for cooperative molecular motor ensembles with tether slack. This is less directly related to biology but provides a mathematical language for understanding constrained movement, which can be useful when modeling how biological components interact under physical limitations.

The work that matters most is the framework for outbreak analytics based on probability generating functions, which offers a systematic computational methodology for describing epidemiological time series like cases or hospital admissions over time. This approach allows for efficient evaluation of joint distributions by transforming probability generating functions into cumulant generating functions, providing a significant computational advantage over simulation-based methods and paving the way for broader adoption in computational epidemiology.

This mathematical tool is complemented by efforts to improve molecular language models, such as STAMP, which introduces a topology-aware representation that explicitly connects fragment-level learning with precise structural editing. STAMP encodes motif identity, attachment sites, and tree topology separately, allowing learned representations to capture motif similarity and recurring transformations while enabling scaffold-preserving optimization. This is significant because it establishes an executable molecular language where learned chemical units remain directly operable during design.

The work on evaluating large language models in ophthalmology also provides important context for how we assess complex systems; specifically, SIGHT introduces a methodological appraisal framework for this purpose. This framework helps structure the evaluation of these models, which is crucial as these models become more integrated into clinical settings. This assessment effort connects to the broader challenge of modeling complex systems, as seen in research exploring outbreak dynamics and population vulnerability in stochastic epidemic models on networks.

The most significant finding today relates to understanding the limits of neuropsychiatric classification, which is crucial because it touches upon how much information a neural network can actually hold before its predictions become meaningless. This idea is explored in Energy, Information, and the Limits of Neuropsychiatric Classification /papers/2610.14095. The work suggests that there are inherent limits to the information processing capabilities within these systems when dealing with complex psychiatric states.

This limitation is further illuminated by examining how information behaves in recurrent networks when subjected to realistic noise, feedback, and nonlinearities, as detailed in Information in a recurrent Retina-V1 network with realistic noise, feedback and nonlinearities /papers/2603.27347. This study looked at the information flow within these specific neural architectures under challenging conditions.

A more practical application of this kind of modeling comes from EEGDash, which is an open-source platform for machine learning on public neurophysiological data /papers/2606.16041. This platform allows researchers to test these complex theories using real, publicly available brain data.

We also see related work in predicting cognitive tasks from single-channel EEG through task-semantic alignment predicting cognitive tasks from single-channel EEG through task-semantic alignment, which attempts to map external task demands onto neural signals for prediction. This connects with the structural plasticity explored in Rewiring of adult neural networks - learning from structural plasticity across multiple scales /papers/2610.18024, which investigates how the physical structure of adult brain networks changes over time based on learning experiences.

The work on modeling delayed biological processes matters because it offers a way to create more flexible and computationally efficient models for systems where events don't happen instantly. The hypoexponential generalization of transit compartment models allows researchers to estimate the mean and variance of delay distributions directly as real-valued parameters, which avoids the artificial constraints imposed by requiring an integer number of compartments in traditional transit compartment formulations. This approach is useful for applying these models across various biological scenarios, such as viral eclipse phases or lymphocyte proliferation.

This efficiency in modeling delays connects to the work on neurogliovascular unit activity, where a simple model of this activity with noradrenaline leading force was developed to understand how these units function. Furthermore, the sense-making phase transition framework for affect provides a conceptual lens for understanding how complex emotional states emerge from underlying processes. While these frameworks offer different perspectives, they all contribute to building richer predictive tools in complex biological systems.

The characterization of fungal infections in the All of Us Research Program highlights a critical issue: the quality and completeness of clinical data when studying rare or under-reported conditions like aspergillosis. This finding underscores that even valuable large datasets require careful assessment regarding missing organism-specific tests to ensure robust clinical conclusions about disease burden.

Today's papers

The papers

Important terms

Thermodynamic Consistency
This is about ensuring that the mathematical optimization process for biochemical reaction networks stays consistent with fundamental thermodynamic laws. It provides a rigorous way to understand how these systems truly behave under different conditions.
Bayesian Uncertainty Quantification
This technique measures how uncertain an AI's prediction is, especially when the labels it uses depend on who is reading or what annotation rules are applied. It helps gauge the reliability of AI outputs in complex scenarios.
Structure-Preserving Discretizations
EPITIME uses this method to model complex epidemic dynamics by keeping the original structure of the system intact during discretization. This allows for more accurate modeling of disease spread.
Pairwise Information Flow
This is a computational method used to reconstruct how different parts of a complex nonlinear network are connected. It helps map out network relationships without being overwhelmed by too much data.