ContextTAD: Context-aware boundary learning for TAD calling from Hi-C contact maps [new] produces asm-optimized left/right boundary tracks via multiscale feat. extraction from broader Hi-C windows, with novel pair & count objectives.

Phylogenetic Tree Inference with Tropical Axial Attention [new] ...uses max-plus operators for piecewise-linear distance learning, leveraging tropical Grassmannian geometry for tree inference.

Attention-Based Multimodal Survival Prediction with Cross-Modal Bilinear Fusion [new] ...integrates histology, RNA, and clinical data to model conditional interactions for patient survival.

Frequency-Space Mechanics: A Sequence and Coordinate-Free Representation for Protein Function Prediction [new] ...encodes proteins as mechanical harmonics graphs (MHGs) from vibrational modes and couplings, using physics alone to predict molecular function.

ToolMol: Evolutionary Agentic Framework for Multi-objective Drug Discovery [new] It merges a multi-objective genetic algorithm with an LLM agent, using a RDKit toolbox for precise ligand modification and chemical knowledge exploitation.

ENSEMBITS: an alphabet of protein conformational ensembles [new] defines the first tokenizer for protein conformational ensembles, providing a discrete vocabulary to integrate protein dynamics into language modeling and design.

MetaGEM: Bottom-Up Reconstruction of Genome-Scale Metabolic Networks via Deep Enzyme-Metabolite Anchoring [new] ...uses multimodal AI combining protein LM with 3D metabolite representations to predict enzyme-metabolite interactions, building functional GEMs.

AI-guided design of efficient perovskite solar cells operationally stable at 100°C [new] ...focusing on material composition and interface engineering.

BioByte 159: the Release of OpenBind, uPAR CAR-T Breaks Down Tumor Defenses, Triplet Tumbling Microscopy Improves Detection of PPIs, and the Potential Role of Autoantibodies in Schizophrenia [blog]

CardioSafe: Multi-task prediction of cardiac ion channel activity with reverse-leak audited benchmarking [new] Integrates chem/Tx features to predict multi-ch. cardiac ion activity; audit found widespread data overlap in comps.

ENSEMBITS: an alphabet of protein conformational ensembles [new] The first tokenizer for protein conformational ensembles, enabling dynamics in language models and design, even predicting from single structures.

ToolMol: Evolutionary Agentic Framework for Multi-objective Drug Discovery [new] Combines a multi-objective genetic algorithm with an agentic LLM operator, using RDKit-backed functions to iteratively update ligand populations for drug design.

Hessian Matching for Machine-Learned Coarse-Grained Molecular Dynamics [new] ...augments force matching with stochastic second-order curvature data to constrain CG free-energy surface shape, creating more accurate potentials.

Multitask Multimodal Fusion with Tabular Foundation Models for Peak and Durability Prediction of Pertussis Booster Response [new] predicts distinct peak, durable pertussis booster resp. via a multi-task contrastive multimodal fusion arch. for hetero. data.

Chem-GMNet: A Sphere-Native Geometric Transformer for Molecular Property Prediction [new] ...replaces standard transformer modules with sphere-native counterparts, processing molecular data via spherical embeddings, recurrence, and kernel operations.

Predicting Endocrine Disruptors: A Deep Learning QSAR Model for Estrogen Receptor Activity [new] ... which uses a DNN to rapidly identify ER-binding molecules for risk assessment and conservation.

Engineering proteins with Sequence Display [new] enabling unbiased, large-scale mapping of sequence-activity landscapes, providing rich datasets to guide machine learning.

Deep peptide recognition profiling decodes TCR specificity and enables disease-associated antigen discovery [new] predicts T cell receptor specificity and identifies disease-associated antigens.

A generative artificial intelligence approach for peptide antibiotic optimization [new] ...redesigns them for enhanced efficacy against drug-resistant bacteria, validated through lab tests and mouse infection models.

Integrating Diffusion and Liquid AI Models for Predicting Peptide Affinity from mRNA Display Selections [new] predicts binding affinity (ΔG°) for exp. & generated peptides, enhancing ligand discovery and mol. property prediction.

Set-Aggregated Genome Embeddings for Microbiome Abundance Prediction [new] Predicts microbiome abundance by aggregating genomic language model embeddings from raw DNA sequences, enabling few-shot learning.

Letting the neural code speak: Automated characterization of monkey visual neurons through human language [new] A closed-loop system with digital twins describes V1/V4 selectivity using semantic hypotheses generated from high/low-activating images.

SCOPE: Siamese Contrastive Operon Pair Embeddings for Functional Sequence Representation and Classification [new] employs PLM embeddings & Siamese MLP for scalable operon pair class. across diverse microbial genomes, aiding annot. & network reconstruction.

Learning the Interaction Prior for Protein-Protein Interaction Prediction: A Model-Agnostic Approach [new] by designing a bio-informed classifier based on L3 rule, using graph prompt learning to reformulate embed pair classification into a graph task.

Clin-JEPA: A Multi-Phase Co-Training Framework for Joint-Embedding Predictive Pretraining on EHR Patient Trajectories [new] enables stable co-training encoder & predictor for patient traj. forecast & diverse risk pred. from single backbone w/o fine-tuning.

Deep Learning for Protein Complex Prediction and Design [new] ...investigates domain-specific architectures and search algorithms to model hierarchical protein structures, identify interacting homologs, and design protein sequences.

Beyond Manual Curation: Augmenting Targeted Protein Degradation Databases via Agentic Literature Extraction Workflows [new] introduces an AI workflow to extract structured TPD records, including critical context, from lit, expanding existing databases.

Unlocking LLM Creativity in Science through Analogical Reasoning [new] leverages cross-domain analogies to combat mode collapse, yielding diverse and novel scientific solutions for autonomous discovery.

LPDP: Inference-Time Reward Control for Variable-Length DNA Generation with Edit Flows [new] Training-free local re-solving for variable-length DNA generation, guiding inference by evaluating edit actions via bounded discrete programs.

Self-organized MT Direction Maps Emerge from Spatiotemporal Contrastive Optimization [new] ...revealing a discriminative pressure vs. spatial regularization trade-off drives their formation, unifying cortical self-organization principles.

Decomposing the Generalization Gap in PROTAC Activity Prediction: Variance Attribution and the Inter-Laboratory Ceiling [new] ...reveals inter-lab measurement variance dominates the generalization gap for novel-target pred., setting an inter-lab ceiling.

Five burning Qs for Isomorphic Labs after $2.1B raise for AI biotech [blog] cementing its bellwether role in using AI for drug discovery with a record-setting private biotech raise.

Reconstructing True 3D Spatial Omics at Single-Cell Resolution [updated] ...by modeling tissue evolution as a continuous dynamic vector field, extracting uninterrupted, infinitely resolvable states from discrete slices.

TD3B: Transition-Directed Discrete Diffusion for Allosteric Binder Generation [new] ...generates allosteric binders with specified agonist or antagonist behavior by controlling protein state transitions.

Cortico-cerebellar modularity as an architectural inductive bias for efficient temporal learning [new] This modularity, with a cerebellar module driving efficiency, enables the recurrent core to function as a fixed reservoir for temporal learning.

Clin-JEPA: A Multi-Phase Co-Training Framework for Joint-Embedding Predictive Pretraining on EHR Patient Trajectories [new] stably co-trains encoder & latent trajec. pred. via 5-phase curr. for fine-tune-free patient traj. forecasting & diverse risk pred.

AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents [new] enables in silico phenotypic screen prediction across 1,920 CRISPR screens, formulating it as gene rank prediction for virtual cell progress.

Yeti: A compact protein structure tokenizer for reconstruction and multi-modal generation [new] It discretizes continuous atomic coordinates into discrete tokens for scalable multimodal protein sequence/struct co-generation, enabling unified repr. and design.

FLUX: Geometry-Aware Longitudinal Flow Matching with Mixture of Experts [new] FLUX: Geometry-Aware Longitudinal Flow Matching with Mixture of Experts identifies regimes and models geometric transport in unpaired data through expert decomposition.

Pushing Biomolecular Utility-Diversity Frontiers with Supergroup Relative Policy Optimization [new] It directly constructs rewards from set-level diversity, ensuring balanced utility and broader biomolecular generation beyond narrow candidates.

MicroFuse: Protein-to-Genome Expert Fusion for Microbial Operon Reasoning [new] Integrates protein identity and genomic context using specialized experts to resolve agreement and conflict, for operon co-membership prediction.

A putative, computationally stable structure of homotrimeric BP180/collagen XVII [new] ...was predicted and validated via molecular dynamics, offering a foundation for pharmacological targeting and understanding bullous disease epitopes.

Meow-Omni 1: A Multimodal Large Language Model for Feline Ethology [new] ...natively fuses video, audio, physiological data, and text to infer feline intent, overcoming semantic aliasing for inter-species understanding.

How Much is Brain Data Worth for Machine Learning? [new] Quantifies its value via scaling laws, exchange rates with task samples, and identifies conditions for robustness gains and when its collection is worthwhile.

LiteMedCoT-VL: Parameter-Efficient Adaptation for Medical Visual Question Answering [new] Adapts compact medical VQA models by transferring multi-step reasoning from large models using LoRA fine-tuning on explanation-enriched data.

Predictive and feedback signals differently shape the formation of group-level and individualized language representations [new] predictive signals drive common neural learning architectures, as feedback mechanisms explain differences in language acq.

Learning the Interaction Prior for Protein-Protein Interaction Prediction: A Model-Agnostic Approach [new] ...learns this prior with a classifier leveraging the 'L3 rule,' generating L3-path-regularized prompt graphs from embeddings for graph classification.

Towards the explainability of protein language models [new] by surveying explainable AI methods applicable to these models.

Machine learning cross-platform proteomic imputation enables protein quality scoring and replication of epidemiological associations [new]

PPI-Net connects molecular protein interactions to functional processes in disease [new] Models disease propagation integrating PPI networks with pathway hierarchies using hierarchical GNN, mapping gene signals to functional programs for mechanistic insight.