Benchmarking AlphaFold and related deep learning approaches for modeling antibody and TCR antigen recognition [new] evaluates DL protocols on diverse ICx, revealing varied improv. by interface type, pooling boosts success.

Diffusion learning reveals viable parameter manifolds and compensation geometry in biological dynamical systems [new] identifies parameter sets producing target behaviors by inverting feature maps, uncovering compensation and hidden parameter dependencies.

Folding, Reasoning, and Scaling with Open-source Drug Discovery Engine [new] ...is achieved through an all-atom biomolecular foundation model using co-folding as a shared structural reasoning layer for drug design and optimization.

Spectral Diffusion for Protein Dynamics [new] applies Fourier transforms in the spectral domain to disentangle slow/fast dynamics modes, enabling rapid prediction across temperatures.

Predicting Therapeutic Outcome via Aligning Patient-Specific Knowledge Graph and Gene-Level Perturbation Representations [new] to predict drug response via a multiview framework aligning personal mechanistic embeddings & simulated transcriptomic profiles.

Score Distributions, Not Cells: Evaluating Single-Cell Perturbations Under Class Overlap [new] New metric (CDS) averages per-cell probabilities into population profiles, overcoming class overlap to reliably identify perturbations.

Data-Driven Soft Labeling Scales DNA Read Classification to Whole-Body Cell-Type Deconvolution [new] ...by estimating conditional cell-type distributions for individual DNA reads, resolving label conflicts & non-discriminative signals across many cell types.

Rethinking Benchmarks and Models for Enzyme Specificity Prediction [new] reveals current models struggle on discovery-relevant tasks; interaction-aware representations from full complexes show promise for enzyme prioritization.

PEPstrMOD2: Next-generation tertiary structure prediction of chemically modified and non-natural peptides [new] DL integration facilitates init. str. pred., exp. chem space signif., & remov. len. restr. for complex peptides.

DELPHAI predicts heterogeneous perturbation responses with learned cell fitness and gene-space retrieval [new] filtering depleted cells & bypassing latent bottlenecks, enabling explainable CT-spec survival insights w/o bio priors.

BoltzMol-1: Towards Reliable Virtual Screening for Fast and Cost-Effective Hit Discovery [new] This pipeline couples model-driven ranking with early ADMET filtering to prioritize developable hits across diverse target classes.

A Conversation with Jack Scannell: The Intractable ‘Sociotechnical’ Problems of Drug Discovery and Development, Genetic Triumphalism and Predictive Validity [blog]

Water as a thermal contrast agent for artificial-intelligence-enhanced in vivo mid-infrared thermography [new] This approach provides high-res, hi-contrast in vivo imag. of vasc., tumors, & mets, improving early diag. & surg. prec.

CatESO: Differentiable Enzyme Sequence Optimization Guided by Substrate-Aware kcat Prediction [new] Designs substrate-specific catalytic turnover using gradient-guided backprop through a kcat pred, integrating plausibility & struct.

A control-validated pan-proteome deep-learning pipeline nominates GPR35 as a candidate target of the orphan bacterial metabolite ligiamycin A [new]

Guiding generative models to uncover diverse and novel crystals via reinforcement learning [new] ...to find thermodynamically stable, diverse crystalline materials with desired properties.

How to design a cancer vaccine (and vastly improve them): Alex Rubinsteyn & Ben Vincent [blog] by precisely selecting tumor-spec. neoantigens & CTAs via adv. comp. & immunopeptidomics for effective, personalized immunotherapy.

Benchmarking large language models for ACMG/AMP variant interpretation and variant calling [new] using a novel framework assessing safety, prov., & reproducibility, attributing failures to spec. arch. layers w/ temp. blinded data.

Tissue tearing degrades optimal-transport and diffeomorphic registration of spatial transcriptomics beyond displacement magnitude: a multi-seed deformation benchmark and a supervised graph cross-atten... [new]

Real-time mass defect-driven prediction of glycopeptide precursors enables enrichment-free serum glycoproteomics [new] ... by using machine learning for direct N-glycopeptide classification in DDA from serum.

Deciphering the Tumor Microenvironment: An Integrated Single-Cell RNA-Seq and AI Framework for Novel Biomarker and Therapeutic Target Discovery in Melanoma [new]

ADMET Property Prediction with Quantum-Inspired Preprocessing [new] Encodes molecular descriptor dependencies into a many-body H, using simulated time evolution to generate compact, interpretable features for ADMET property modeling.

Task-adapted biological foundation models uncover perturbation-centric representations [new] Supervised perturbation transforms bio lat. space, aligning same-pert. responses, capturing links.

Improving Generalizability in Whole-Cell Antibiotic Discovery Through Active Learning [new] by strategically balancing compound novelty and predicted bioactivity for OOD extrap., efficiently discovering hits in whole-cell screens.

TRIOPS: A deep learning framework for prediction of T cell receptor-MHC binding specificity [new] Predicts TCR-MHC restrict. from seq, extends to unseen alleles via cross-react. sampling, identifies allele-spec. HLA loss in tumors.

PRISM : Peptide-specificity annotation of T-cell receptors with uncertainty quantification [new] maps TCRs to a peptide-organized latent space via metric-learning, augmenting training with synthetic data to enable OOD abstention.

VirProtRAG: Literature-grounded viral protein function annotation with retrieval-augmented generation [new] Uses hybrid retrieval & evidence reranking for verifiable annotations, expanding and refining viral protein characterization.

Accurate, comprehensive gene annotation and ortholog identification across thousands of vertebrate genomes with TOGA2 [new] Uses exon orthology, DL, & gene tree recon creates comprehensive comparative G res. for thousands of genomes.

A High-Confidence Atlas of Protein Methylation Enables AI-Driven Detection of Methylated Peptides [new] ...by creating a stringent human methylation atlas from reanalyzed public data to train a DL model for improved peptide detection

Unbalanced Perturbation Dynamics For Cell Fate Design [new] models transcriptomic and cell-number dynamics from single-cell snapshots for predicting responses and designing interventions to achieve desired cell fates.

Generative cell phenotyping with structured latent populations [new] ...is achieved via a semi-supervised VAE with a Gaussian mixture prior, unifying cell classification, discovery, density estimation, and anomaly detection.

Context-dependent calibration of Evo2 likelihood with bacterial fitness: a quantitative characterization across five E. coli datasets [new]

Strict OOD Antigen-to-Antibody Retrieval with CDR-Aware Slot Late Interaction [new] ...focuses on robustly retrieving antibodies for unseen antigens by constraining antibody latent slots to CDRs and scoring local slot compatibility.

SAPPTree: Identification of an S-Acylation Motif Drives a Novel S-Acylation Prediction Program [new] Uses novel S-acylation motif (cys cluster, hydro stretch, polybasic helix) & struct. features in a RF model to predict S-acylation.

MycoCirc: A Pan-Fungal Multi-Modal Pretrained Model for Fungal circRNA Prediction from Genome Sequence [new] integrates 3 modalities via cross-attn to predict fungal circRNAs from genomes, offers a comprehensive prediction atlas.

Folding scFv--Antigen Complexes at Scale [new] by generating large cofolded structure ensembles to characterize model reliability and the impact of diverse inference-time settings.

Pathogen context reshapes antimicrobial peptide generation [new] by using genome features and strain descriptions to condition peptide generators for target-specific libraries.

CodonBERT and ESM-2 Embedding Spaces Share an Evolutionarily Conserved Paired Geometry Encoding Synonymous Codon Information [new] ...arises from indep trained nuc/prot models converging on shared, lin retrievable & evol constr. reps

Distillation enables scalable high-fidelity virtual screening across ultra-large chemical libraries [new] Transferring structure-based model's predictive power to efficient seq-based surrogate, trained on small frac. of large lib.

Benchmarking the translational potential of AI-based drug-resistance prediction from Mycobacterium tuberculosis whole-genome sequencing data [new]

Structural Organization of the Nvj3-Mdm1 Complex Reveals a Conserved Lipid-Compatible Contact Site Module [new] Module's heterodimer forms lipid-comp. conduit at the interface, suggests direct interorganelle lipid xfer at the NVJ.

Unveiling Cerebrospinal Fluid Protein Biomarkers in Pediatric Acute Lymphoblastic Leukemia Using Proximity Extension Assay [new] reveals distinct CSF protein sigs for ALL disease effect, CNS inv, and monitoring treatment response.

A Comprehensive Evaluation of Protein Structure Prediction Models for Short Peptides [new] Studies DL models' handling of short peptides, observing variation by length & secondary structure. Consensus helps define struct. hypotheses.

GenPerturb: sequence-grounded interpretation of perturbation transcriptomes using pretrained genomic models [new] Links Perturb-seq gene responses to cis-reg elements & TF motifs by sequence analysis, without matched chromatin data.

Raw-count embeddings improve single-cell foundation models [new] by finding simple log-transformed raw counts suffice, rendering complex normalisation, gene ordering, and positional encoding unnecessary.

A generalizable interface-seeded framework for de novo design of functional oligomers [new] ... designs modular responsive homo-oligomers for conditional/reversible assembly, building ligand-dependent biosensors & gene switches.

Using spIsoNet to address the preferred-orientation problem in cryoEM reconstructions [new] ...through deep learning for anisotropic map correction and enhanced particle pose estimation in cryoEM data.

AART enables fast and accurate cross-platform proteomic translation [new] ...via integrated regression and residual learning, addressing technical differences to enable reproducible and integrated proteomic research across platforms.

Evo 2's Perception of Single Nucleotide Substitutions in the Genes of Two Plant Model Organisms [new] ...perceives simulated nucleotide substitutions less likely than observed ones in plant genes, reflecting evolutionary constraints.

Inferring viral proteins that act as public goods during coinfection [new] ...quantitative framework analyzes MS2 evolution data under varying MOI to infer protein-specific public goods; lysis protein is the strongest candidate.