AMARO v2.0 is an AI-powered simulation method that focuses on heavier atoms, not every atom, to study how proteins and drugs move and interact faster. Faster simulations could help researchers understand biological processes and speed drug discovery.
A unified coarse-grained neural network potential for the dynamics of biomolecular complexes · Antonio Mirarchi, Stephen E. Farr, Adrià Peréz et al. · arXiv:2610.37606
An AI method estimates a tissue measure, including myelin content called MPF, from standard hospital MRI scans. It lets researchers study myelin health in diseases like multiple sclerosis using existing patient data.
Retrospective macromolecular pool fraction mapping from conventional clinical MRI: a physics-guided self-supervised deep learning framework · Jelmer van Lune, Miha Fuderer, Oscar van der Heide et al. · arXiv:2610.37213
The method simulates how a patient's heart valve deforms and closes, including tissue contact, much faster using graphics cards. Faster simulations could help doctors understand individual valve function and plan personalized treatments.
A Differentiable, GPU-Accelerated Finite-Element Formulation in JAX for Large-Deformation Shells with Contact: Application to Patient-Specific Aortic Valve Simulation · Mohsen Nakhaei, Alison M. Pouch, Matthew A. Jolley et al. · arXiv:2610.38075
A method uses an AI model and real plus simulated X-ray views to make clearer 3D medical images from few scans. It could let patients get high-quality scans with significantly less radiation exposure.
P^2-Gaussian: Prior-Driven and Projection-Guided Gaussian Splatting for Sparse-View Cone-Beam CT Reconstruction · Jinqiu Xia, Jianhui Ma, Bin Ouyang et al. · arXiv:2610.39359
A test checks whether AI agents follow complex lab instructions for DNA sequencing structures and finds they often miss exact critical details. It highlights a challenge for AI to reliably assist in real-world molecular science.
Evaluating and improving AI agents for sequencing protocol understanding · Chi-Lam Poon, Michael Goneos, Zezheng Song et al. · arXiv:2610.37851
ReactNexus uses generative AI to discover chemical reaction pathways and networks from starting molecules alone, without knowing final products. It may help scientists understand reactions and design more efficient processes or find compounds faster.
Generative discovery of chemical reaction networks · Yunhong Han, Gaopeng Ren, Yan Liu et al. · arXiv:2610.38236
GlycoWeave predicts how proteins bind to complex sugar molecules called glycans and outperforms general AI models like AlphaFold 3. It may help understand diseases and develop drugs that target protein-sugar interactions.
Accurate structural modelling of protein-glycan recognition with GlycoWeave · Huifeng Zhao, Jintu Zhang, Yifei Xiao et al. · arXiv:2610.37410
The AI model turns lower-quality CBCT scans into higher-quality CT scans using the scanner's physics and raw measurements. It may make AI results more trustworthy for critical uses such as radiotherapy, where precision matters.
EPC-RawDC: Equivariant Physics Constrained 3D Diffusion with Raw Data Consistency for CBCT to CT Image Synthesis · Alzahra Altalib, Chunhui Li, Alessandro Perelli · arXiv:2610.40046
HELICS learns patterns from biobank data to generate realistic synthetic human genomes and can tailor them to ancestry or disease risk. It may give researchers privacy-safe genetic data for studies, especially underrepresented groups and rare diseases.
HELICS: Biobank-scale Conditional Synthetic Genome Generation via Latent Flow Matching · Ahmad Abdel-Azim, Xihong Lin · arXiv:2610.36361
AI analysis finds hidden diversity in sugar-adding machinery across double-stranded DNA viruses, including a giant virus that appears to modify its DNA with a sulfated sugar. It expands understanding of viral biology and suggests these sugar changes may have unrecognized roles in how viruses work and interact with hosts.
AI-driven structural informatics reveals diverse glycosylation machinery across double-stranded DNA viruses · Lan Na · arXiv:2610.39662