deep learning

Impact of Rare and Common Genetic Variation on Cell Type-Specific Gene Expression

Single-cell eQTL mapping in 5.4 million PBMCs from 1,925 individuals with matched WGS finds common-variant eQTLs for 83% and rare-variant signals for 47% of 21,404 genes.

A clinically annotated proteomic and histological resource of 1,000 colorectal cancers

Deep proteomics of 997 treatment-naïve colorectal cancers defines six subgroups, splitting chromosomally unstable tumors into three and revealing a rare adverse subgroup, with H&E slide benchmarks.

Complex enzyme active site scaffolding by iteratively detuned catalytic guidance

CaGE alternates guided structure prediction with sequence redesign, fading guidance each cycle, so sequences recover complex catalytic sites unaided; designed nonheme-iron and PLP-dependent enzymes were active.

Phylogenetic estimation of diversity-dependent biogeographic rates using deep learning

DDGeoSSE, a GeoSSE extension in which local species richness modulates speciation, extinction and dispersal, is fit with phyddle neural networks that detect diversity dependence in Anolis and Viburnum.

Learning Dynamic Protein Representations at Scale with Distograms

Adding Boltz2 distogram edges and features to protein and RNA graph networks beats MD-correlation graphs on binding tasks and improves stability, interaction and RNA predictions without simulation.

Anisomycin-induced ribosome collisions follow a dose-dependent positional rule, distinct from the UV-induced collision landscape

Low- and high-dose anisomycin confine ribosome collisions to a dose-shifted zone just past the start codon, whereas 30 min after UV, collisions appear more dispersed and favor highly translated mRNAs.

Human-specific CCG-repeat insertion in the RBFOX1 promoter implicated in autism and speech enhances EGR1-responsive activity

A human-specific CCG insertion in the RBFOX1 promoter boosts EGR1-driven reporter activity in HEK293T cells, as does an autism-linked CCG-gaining duplication in PTCHD1.

Large scale prospective evaluation of co-folding across hundreds of Mac1-ligand complexes and three virtual screens

On 489 evaluable new Mac1 complexes, AlphaFold3, Chai-1 and Boltz-2 placed over half of ligands within 2 Å, but on docking hit-lists AF3 and Boltz-2 added little over docking.

AlphaFold3 and Intrinsically Disordered Proteins: Reliable Monomer Prediction, Unpredictable Multimer Performance

AlphaFold3 pLDDT matches AlphaFold2 at calling disordered residues (Matthews correlation 0.693 vs 0.701); on disorder-containing complexes mean DockQ is similar, but interface features explain less variance (42% vs 58%).

Neural prediction decorrelation reveals that adversarial robustness substantially improves DNN prediction accuracy across the entire human auditory cortex

Sounds synthesized to decorrelate model predictions across human auditory cortex show an adversarially robust CochResNet50 predicts fMRI responses far better than its standard counterpart, a gap hidden with natural sounds.

Illuminating dark genes and dark metabolites: calibrated functional association across kingdoms with autoencoders and correlation

VaLPAS pairs a bi-directional autoencoder with correlation for omics guilt-by-association with calibrated confidence; in L. starkeyi its autoencoder linked 176 unidentified metabolites to known-function proteins at ≥50% confidence.

Joint protein, mRNA and DNA sequence design and optimization with nucleotide-level Potts models

A structure-conditioned nucleotide-level Potts model designs protein and coding sequences jointly, enabling tRNA adaptation and RNA motif embedding beyond synonymous codons with designability comparable to Caliby.

PIE: Generalizing perturbation effects across unseen perturbations, contexts and datasets

Predicting differentially expressed genes and expression shifts per condition from database embeddings plus per-gene evidence, PIE reaches 1.2–3.2 times the best baseline's AUPRC on unseen contexts, perturbations and datasets.

LGTM: Gaussian Process Modulated Neural Topic Modeling for Longitudinal Microbiome

LGTM couples a neural topic model with additive basis-function Gaussian processes, learning microbial co-abundance topics and how covariates relate to their trajectories in longitudinal gut microbiome data.

Ref(2)P/p62 protein pleomorphism confers tissue-specific function

One Drosophila Ref(2)P polypeptide takes different structures in midgut and muscle, driving autophagy or mitochondria-associated degradation; deleting disordered region IDR2 switches muscle Ref(2)P to midgut state.

Reliable Evaluation of Transductive Population Graph Neural Networks for Multisite Autism fMRI

Site-aware population GNN on ABIDE-I: cohort AUC about 0.94 (kept with site-only edges) was not reproduced by a site-prior classifier (0.506); a directed-edge support gain (+0.252) vanished when symmetrised (−0.003).

Accurate ab initio gene prediction in eukaryotes with Tiberius in multiple clades

Extended Tiberius models outscore Helixer and ANNEVO on all 33 genomes from the five clades all three tools cover, and approach BRAKER3 accuracy in four clades without extrinsic evidence.

Physics and Morphology Constrained Quantitative Susceptibility Based Segmentation of Cerebral Veins

Dipole-field physics and Frangi vesselness losses lift QSM vein segmentation to centreline Dice 0.70 (U-Net 0.57); derived venous OEF tracks calibrated fMRI better than vessel filtering.

CycloCross: Adversarial single-cell RNA-seq data translation across species

Translating cells between mouse, macaque and human scRNA-seq with a CycleGAN, without ortholog mapping, preserved developmental time structure, imputed missing states and produced more realistic synthetic cells than scGAN.

TxCyto: A machine learning framework for estimating cytokine activity from whole transcriptome

A fully connected neural network trained on 9491 TCGA bulk tumor transcriptomes estimates activity of 645 cytokines and secreted proteins, checked against perturbation and protein data.

A Taxonomy-Informed Sparse DNA Foundation Model for Microbial Genomics

A sparse mixture-of-experts DNA model conditioned on hyperbolic taxonomy embeddings learns better microbial representations than its taxonomy-free twin and comes close to Evo2-7B in probing and fine-tuning at lower compute.

Accessing Enzyme Kinetic Data and Prediction Methods at Scale

OpenKinetics Predictor unifies thirteen enzyme kinetic predictors behind one interface and API; on E. coli iML1515, eight kcat predictors correlate pairwise at r 0.08 to 0.67.

An RNA Language Model trained on sequence alone reveals the structural logic of Internal Ribosome Entry Sites

Continued pretraining of an RNA language model on picornaviral 5′ UTRs alone predicts IRES base pairs with median precision 0.80 versus 0.47 for Infernal, and reveals a Type II subclass.

BOTANIC-1: a series of long-context plant genomic foundation models in the agentic era

Bidirectional Mamba2 plant genomic language models (318M to 3.2B parameters) trained on 320 land-plant genomes top a 22-task frozen-model aggregate; a small variant reaches 131,072 bp context.

TCF25 emerges as a cytoplasmic regulator of GPRASP2 stability

TCF25 is mostly cytoplasmic with a perinuclear membrane-proximal pool, stabilises GPRASP2 against proteasomal degradation, and its C-terminal residues 604-664 are required for their association.

BFVD v3-UniProt-complete, improved viral protein structure predictions

5,776,417 ColabFold-AlphaFold2 structures cover 98.9% of viral UniProt 2025_03 sequences, 16.4-fold more than earlier versions, with 75.3% high-confidence and 72.6% of viral reference proteomes fully covered.

BoneGraph: A Domain-Specialised, Self-Correcting Reasoning System for Bone Science Retrieval, Grounded Inference, and Image Mechanics

BoneGraph combines a SPECTER2-indexed bone corpus, knowledge graph, deterministic physics checks, and user-taught rules in a local five-tab system; oracle passages raise accuracy 42% to 78%.

Automatic Quality Control and Error Correction in MRI linear registration via a Residual Parameter Prediction Network for T1w MRI

A 3D ResNet18 trained on synthetically misaligned T1w scans predicts residual affine parameters, correcting synthetic misalignments to 0.778 mm residual RMSE and flagging failed registrations better than RegQCNET or DARQ.

The Unreasonable Effectiveness of Cell Types in Describing Neuronal Physiological Features

On 495 human Patch-seq neurons, cell type means predicted electrophysiology better than neural networks on scGPT embeddings or gene panels; fusing scGPT embeddings with cell type labels predicted best.

Whole-body 3D kinematics of freely behaving Drosophila

Seven high-speed cameras, a hybrid 2D/3D network and body-model inverse kinematics track 50 keypoints on freely moving flies; height and speed peak together at all running speeds.