variational autoencoder

Deciphering Cell Cycle Dynamics and Cell States in Single-cell RNA-seq data with SPAE

This paper introduces SPAE, a model for analyzing cell cycle dynamics and states in scRNA-seq data, showing improved accuracy and robustness in cell cycle characterization and cancer cell cycle transition prediction.

UDIST: unsupervised disentanglement of shape and texture for multi-scale phenotypic profiling in 2D microscopy

UDIST separates shape and texture in microscopy images into independent latent subspaces, enabling unbiased phenotypic analysis across biological scales.

VINE: Variational inference for scalable Bayesian reconstruction of species and cell-lineage phylogenies

This paper introduces a variational inference method for Bayesian phylogenetic reconstruction, showing faster speeds than traditional Bayesian methods on real-world datasets.

Thousandfold Expansion Microscopy

A four-network sodium acrylate/dimethylacrylamide hydrogel, cast ionic-within-ionic without neutral re-embedding, reaches about 1000-fold linear expansion and resolves individual peptide residues on standard confocal microscopes.

UniFlow: Unifying protein conformational ensemble generation and machine-learned force fields with a scalable normalizing Flow

A normalizing flow over backbone internal-coordinate angles generates protein ensembles comparable to BioEmu in one step and doubles as a differentiable coarse-grained potential for Langevin dynamics.

Evidence of predictive information compression in latent space in humans during speech listening

This study compares speech processing models to EEG data, finding that predictive information encoding in latent space better explains neural activity than compression or prediction alone.

Spatial Topology Reveals Biologically Distinct Recurrent Motifs in Colorectal Cancer

The paper introduces STORM, a method to identify recurrent spatial motifs in colorectal cancer without cell-type labels, finding a fibroblast barrier linked to immune exclusion and prognosis.

Quantum Encoding Strategies for Drug Response Prediction: An Exhaustive Benchmark on a 20-Qubit Superconducting QPU

The paper benchmarks twelve quantum encoding strategies for drug response prediction on a 20-qubit QPU, finding that a QAOA-inspired encoding achieves the best RMSE.

Learning proteomic disease trajectories with flow matching

This paper introduces proteome velocity, a framework using flow matching to infer continuous proteome trajectories from proteomics data, showing its application in mouse sepsis and clinical COVID-19.

Dynamics of ML-based Morphological Features Indicate a Shear Stress-Dependent Bifurcation of hiPSC-Derived Endothelial Cell States

The paper uses ML to analyze morphological dynamics of hiPSC-derived endothelial cells under shear stress, identifying a bifurcation between two stable cell states.

Variational Autoencoder-enabled High-throughput Drug Screening for HIV Latency Modulators predicted through Noise in Gene Expression

This study uses a variational autoencoder to screen ~175,000 compounds for HIV latency modulators, identifying 16 LRAs and 2 LPAs with a 15.9% experimental hit rate.

OMIDIENT: Multiomics Integration for Cancer by Dirichlet Auto-Encoder Networks

OMIDIENT is a multi-omics integration method that uses Dirichlet auto-encoders to improve clustering, classification, and data reconstruction across cancer types.

Generative cell phenotyping with structured latent populations

MARVIN is a semi-supervised variational autoencoder for cell phenotyping that identifies populations and detects anomalies using a Gaussian mixture prior in the latent space.

A variational sparse Gaussian-process method for detecting spatially variable genes and cellular interactions from spatial transcriptomics

VISGP identifies spatially variable genes and cellular interactions from spatial transcriptomics data, outperforming existing methods in detecting ligand-receptor pairs.

RNAbpFlow: Base pair-augmented SE(3)-flow matching for conditional RNA 3D structure generation

RNAbpFlow generates RNA 3D structures using base pair conditioning and SE(3)-equivariant flow matching, improving topology sampling in benchmarks.

CyStainer: A transformer-based variational autoencoder for robust marker imputation in high-parameter cytometry

CyStainer imputes markers in high-parameter cytometry data without shared markers, outperforming existing methods in panel merging and dataset integration.

CoLa-VAE: A Cell-Cell Communication-Aware Variational Autoencoder for Representation Learning and Expression Denoising

CoLa-VAE improves denoising and representation learning in single-cell data by integrating cell-cell communication information, enhancing downstream analyses like differential communication detection and spatial deconvolution.

ARCADIA Reveals Spatially Dependent Transcriptional Programs through Integration of scRNA-seq and Spatial Proteomics

ARCADIA integrates scRNA-seq and spatial proteomics to reveal spatially dependent transcriptional programs without cell barcode pairing or feature-level correspondence.

BGC-QDR: A Quantum-Assisted Pipeline for Biosynthetic Gene Cluster Discovery and Ranking from Environmental DNA

BGC-QDR is a quantum-assisted pipeline for discovering and ranking biosynthetic gene clusters from eDNA, achieving 0.789 accuracy and 0.835 ROC-AUC using a variational quantum classifier.

MIND: Multimodal Integration with Neighbourhood-aware Distributions

The paper introduces a multi-omics integration framework that learns patient-specific embeddings from incomplete data using a variational autoencoder with a data-driven prior, achieving better performance on downstream tasks.

Deep Learning-Based Reconstruction: Model Comparison for Variable-Density GRAPPA 1H MRSI

This paper compares deep learning models for improving 1H MRSI reconstruction, finding that U-Net outperforms MultiNet in reducing lipid artifacts and improving metabolite map fidelity.

Elucidating the Design Space of Generative Models for Single-Cell Perturbation Prediction

This paper introduces eVAE, a discrete-latent model for single-cell perturbation prediction, achieving state-of-the-art results on multiple metrics and outperforming existing methods in perturbation ranking.

Latent Representations of Early Brain Development: A Multivariate Normative Model of Brain Structure and Behaviour

This paper uses a semi-supervised autoencoder to create multivariate normative models of brain structure and behaviour in children, showing associations between structural deviations and cognitive/behavioural differences.

Learning a chemistry-aware latent space for molecular encoding and generation with a large-scale Transformer Variational Autoencoder

The paper introduces a Transformer-based VAE for molecular encoding and generation, achieving high reconstruction accuracy and a well-structured latent space aligned with chemical similarity.

Dissecting and directing pathology foundation models

This paper introduces PICASSO, a framework that makes pathology foundation models interpretable and controllable by decomposing embeddings into human-interpretable visual concepts, enabling new biological insights and downstream applications.

Temporally overlapping mechanisms diversify clonal B cell responses in vivo.

Single-cell RNA and BCR sequencing of infected mouse spleens shows class switching overlaps clonal expansion, GC-seeding clones also form plasmablasts, and SHM proceeds at constant rate.

Variational autoencoder for interpretable seizure onset phases detection

This paper introduces a variational autoencoder for detecting seizure onset phases in SEEG data, achieving high recall and interpretable latent features.

GeroEngine: Generative single-cell aging trajectories reveal a bidirectionally traversable identity core and direction-specific inflammatory remodeling

GeroEngine reveals aging trajectories show bidirectional identity recovery and direction-specific inflammation, suggesting homeostatic restoration over inflammation suppression.

Comparative Evaluation of Deep Generative Models for Capturing Topological Features in Brain Structural Connectivity

This study compares deep generative models for generating brain structural connectivity data, finding that WGAN-GP provides the most stable performance while VAE and DDPM have application-specific strengths.

Optimal Gene Panel Selection for Targeted Spatial Transcriptomics Experiments

This paper introduces ReconST, a method for optimal gene panel design in spatial transcriptomics, outperforming existing methods in reconstruction accuracy and spatial pattern preservation.