A major challenge by using a novel blend of traditional supercomputers and quantum computers to simulate a protein cutting itself within a massive, realistic biological setting of over a million atoms.
Multiscale Quantum-Classical Simulation of Protein Autoproteolysis in a Million-Atom Membrane Environment · Alexander Orlov, Alexis Ralli, Tim Weaving et al. · arXiv:2610.23108
This paper demonstrates a new way to discover molecules by having AI programs work directly with lab experiments in a continuous loop. Instead of just designing molecules on a computer, AI agents create short DNA/RNA molecules called aptamers, test them in the lab, and then use those real-world results to iteratively refine their next designs.
Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery · Zachary Cutts, John Coller, Rachel Chen et al. · arXiv:2610.24806
A way to get super-sharp images of the retina, which is vital for spotting eye and brain diseases early. Instead of relying on bulky, expensive hardware to correct for eye movements and vision distortions, they've developed a smart computer program that does all the fixing digitally.
High-resolution retinal matrix imaging via computational motion and aberration correction · Aymeric Melt, Léo Puyo, Victor Barolle et al. · arXiv:2610.19245
Cutting-edge AI models for structural biology are good at retrieving information about proteins but struggle with precise geometric calculations, a critical skill for scientific applications.
ProteinGeoBench: Frontier Structural-Biology Agents Are Fluent at Lookup but Lost in Geometry · Matt Turk, Brandon L. Neel, Diana M. Lopez et al. · arXiv:2610.26829
This paper uncovers a hidden pitfall when evaluating AI models that classify breast cancer subtypes across different hospitals or patient cohorts: if the classification categories don't perfectly match, performance comparisons become unreliable.
Label-Space Mismatch Distorts Cross-Cohort Evaluation in Breast Cancer Molecular Subtyping · Suxing Liu, Byungwon Min, Zhijian Zheng et al. · arXiv:2610.24962
Many studies using single-cell RNA sequencing rely on "topic models" to find gene programs, but a common statistical error called "double-dipping" can lead to unreliable findings. Even standard fixes for this problem fail with newer models, causing researchers to mistakenly identify many genes as important.
Reducing double-dipping in single-cell topic models with three-way count splitting · Zheng Qu · arXiv:2610.24130
GpuGEM, a new software that leverages graphics processing units (GPUs) to significantly accelerate the complex calculations required for genome-scale metabolic models.
gpuGEM: GPU-based linear optimisation of genome-scale metabolic models · Farid Zare†, Timothy Hulshof†, Ines Thiele et al. · arXiv:2610.24106
HyperTaxa, a new method that creates "fingerprints" for entire microbial genomes by analyzing their DNA patterns and mapping these to the known microbial family tree. What's is that it can identify microbes even if it hasn't seen their specific genome before, as long as they fit into the broader taxonomic hierarchy.
HyperTaxa: Multiscale Genome Representations for Generalized Microbial Taxonomic Inference · Akash Rao, Sushim Gupta, Vishalini Ramnath et al. · arXiv:2610.25673
This paper gathers leading experts to discuss "Markov state models," a crucial technique for making sense of complex computer simulations of how biomolecules move.
LiveCoMS Roundtable: Discussing Markov state models after two decades of development v1.0 · Daniel M. Zuckerman, Gregory R. Bowman, Amedeo Caflisch et al. · arXiv:2610.24144
A more efficient method for creating accurate PET scans by simultaneously determining both the tracer distribution and how much radiation is absorbed by the body directly from the PET data, eliminating the need for a separate CT scan.
Attenuation corrected activity reconstruction for listmode TOF-PET data · Ahmadreza Rezaei, Suleman Surti, Samuel Matej et al. · arXiv:2610.23655