Bio papers — 2026-10-07

Today's focus is on improving how we predict R genes using homology, which is crucial because it helps us understand gene function in a more automated way. We tried HRPv2, an automated and enhanced method for full-length homology-based R gene prediction. This approach aims to find these genes by comparing sequences to known ones, which is a big step toward understanding how these genes work.

Another piece of work explored how causal analysis can reveal autonomy in models of biological systems. This method looks at complex biological models to see which parts are truly independent, giving us insight into the underlying mechanisms. Building on that idea, we looked at Phylo2Vec, which is a vector representation for binary trees used to encode phylogenetic relationships more effectively.

We also touched upon the relationship between discrete and continuous dynamics in biology, specifically when they align and when they diverge. This helps us decide which mathematical framework is best suited for describing biological processes at different scales. Furthermore, we investigated the identifiability of a linear dynamical reaction-diffusion system with directed interactions using only second-order statistics from a single snapshot.

Finally, we looked at how a linear fitness subspace within protein language models enables sample-efficient directed evolution. This work suggests that by understanding this subspace, we can guide evolutionary changes in proteins much more efficiently than traditional methods.

The work that truly matters is the development of mathematical invariant-enabled topological neural networks for predicting molecular and materials properties, which suggests a new way to model complex physical systems. This approach attempts to capture underlying structural symmetries in data, which could lead to more robust predictions than standard machine learning methods.

This network framework utilizes mathematical invariants to predict properties of molecules and materials, aiming for greater accuracy in material science applications. Furthermore, encoding level-three semi-directed phylogenetic networks using quarnets and quinnets provides a novel way to represent complex evolutionary relationships within biological data. This method offers a more sophisticated structure for analyzing how genetic information is organized across different species.

Another piece of work focuses on retinalysis-vascx, which is an explainable software toolbox designed for extracting retinal vascular biomarkers from images. This tool aims to make the extraction process transparent and understandable, which is crucial for clinical interpretation of eye health data. This contrasts with the database update IMPPAT 3.0, which refines a FAIR database of phytochemicals and formulations from Indian medicinal plants, providing a more accessible repository for ethnobotanical knowledge.

Today's papers

The papers

Speak to a Protein: An Interactive Multimodal Co-Scientist — Building a working mental model of a protein typically requires weeks of reading, crossreferencing crystal and predicted structures, and inspecting ligand complexes, an effort that is slow, unevenly accessible, and often requires specialized computational skills. [episode]
retinalysis-vascx: An explainable software toolbox for the extraction of retinal vascular biomarkers — Automatic extraction of retinal vascular biomarkers from color fundus images (CFI) is crucial for large-scale studies of the retinal vasculature, and this paper presents VascX, an open-source software toolbox that extracts these biomarkers from artery-vein segmentations to provid [episode]
Phylo2Vec: a vector representation for binary trees — Binary phylogenetic trees are fundamental to understanding evolutionary history, but inferring latent nodes in these trees is computationally expensive. [episode]
Discrete vs. continuous dynamics in biology: When do they align and when do they diverge? — Many biological systems are governed by difference equations and exhibit discrete-time dynamics, but this work establishes a mathematical framework to bridge these discrete and continuous representations, showing how they align precisely at discrete times while offering approxima [episode]
How causal analysis can reveal autonomy in models of biological systems — Standard techniques for studying biological systems largely focus on their dynamical, or, more recently, their informational properties, usually taking either a reductionist or holistic perspective. [episode]
Encoding level-3 semi-directed phylogenetic networks by quarnets and quinnets —
Effects of Periodic Migration on Selection in Subdivided Populations —
Neural networks as decision trees: an analytical solution for learning and neural selectivity —
HRPv2: an automated and enhanced method for full-length homology-based R-gene prediction —
A Time-Resolved Framework for Quantifying Neuronal Network State Transitions —
Identifiability of a linear dynamical reaction-diffusion system with directed interactions from the second-order statistics of a single snapshot —
IMPPAT 3.0: An updated FAIR database of phytochemicals and formulations of Indian Medicinal plants —
Linear Fitness Subspace in Protein Language Models Enables Sample-Efficient Directed Evolution —
Mathematical Invariant-Enabled Topological Neural Networks for Molecular and Materials Property Prediction —

Important terms

R gene prediction: This is about automatically finding genes that cause disease by comparing their sequences to known ones, which helps us understand gene function in a more automated way.
Causal analysis: This method looks at complex biological models to figure out which parts are truly independent, giving us deep insight into the underlying mechanisms of biological systems.
Phylo2Vec: This is a new way to represent binary trees used for phylogenetic relationships. It helps encode evolutionary data in a more effective and sophisticated manner.
Topological neural networks: These are new mathematical models that use structural symmetries in data to predict properties of molecules and materials, aiming for more robust predictions than standard machine learning.