contrastive learning

Phylogenetic tree-aware positive-unlabeled deep metric learning for phage-host interaction identification

The paper introduces a phylogenetic tree-aware positive-unlabeled deep metric learning framework to improve phage-host interaction prediction by incorporating biological constraints and handling unlabeled data.

Harnessing the Power of Single-Cell Large Language Models with Parameter Efficient Fine-Tuning using scPEFT

This paper introduces scPEFT, a parameter-efficient fine-tuning framework for scLLMs that improves task adaptation with reduced computational costs and enhances performance in disease-specific and cross-species analyses.

Generative AI Enables Breast Cancer Genomic Subtype Prediction from Histology Images

PATH-IC uses generative AI to predict breast cancer genomic subtypes from histology images, achieving high accuracy and correlating with clinical outcomes.

Contrastive learning for cell division detection and tracking in live cell imaging data

This paper introduces a contrastive learning and graph-based method to improve cell division detection and tracking in live-cell imaging data, achieving better performance across different temporal resolutions.

To Predict is to Design: Unlocking Generative Capabilities in All-Atom Structure Predictors via Geometric Score Distillation

This paper introduces DREAM, a framework that enables protein design by inverting all-atom diffusion predictors, allowing gradient-based optimization of complex modalities like cyclic peptides.

Multiple instance learning on tile level-pathologyimages provides accurate and interpretableclassification for breast cancer molecular subtypes

This paper presents a weakly supervised deep learning approach for breast cancer molecular subtyping using tile-level pathology images without tumor annotations, achieving high accuracy and interpretability.

Diffusion Models vs. DCGANs for Class-Imbalanced Lung Cancer CT Classification: A Comparative Study

This study compares diffusion models and DCGANs for class-imbalanced lung cancer CT classification, finding diffusion models achieve higher accuracy and better recall for benign cases.

DINOcell: Learning Generalizable Perturbation Effects through Self-Distillation

DINOcell predicts cellular responses to perturbations using self-distillation in single-cell transcriptomics, outperforming baselines for single and combination gene effects.

FragmentScope - exploring the fragment space with learned surface representations

FragmentScope uses learned protein surface representations to guide fragment placement and small molecule design, achieving accurate placement and promising lead compounds.

Network-aware self-supervised learning enables high-content phenotypic screening for genetic modifiers of neuronal activity dynamics

The paper introduces Plexus, a self-supervised model for capturing neuronal activity dynamics, and demonstrates its effectiveness in identifying genetic modifiers of neuronal activity in frontotemporal dementia.

High-Parameter Spatial Multi-Omics through Histology-Anchored Integration

This paper introduces SpatialEx and SpatialEx+ frameworks to integrate spatial multi-omics data using histology as an anchor, achieving high-parameter co-profiling across tissue sections.

scCotag: Diagonal integration of single-cell multi-omics data via prior-informed co-optimal transport and regularized barycentric mapping

scCotag improves single-cell multi-omics integration by using co-optimal transport and regularized barycentric mapping, outperforming existing methods in alignment and embedding accuracy.

OS2CR-Diff: A Self-Refining Diffusion Framework for CD8 Imputation from One-Step Inference to Conditional Representation

OS2CR-Diff improves CD8 biomarker imputation in mIF images by using a self-refining diffusion model with multiple conditional inputs and a feature fusion module.

Enhancing STED Microscopy via Fluorescence Lifetime Unmixing and Filtering in Two-Species SPLIT-STED

This paper introduces a method to improve STED microscopy by using fluorescence lifetime unmixing and filtering to enhance resolution and accuracy in two-species imaging.

Multi-modal, multi-species, and multi-task latent-space model for decoding level of consciousness

This paper introduces a multi-modal, multi-species, and multi-task latent-space model to decode consciousness levels from neural recordings, achieving cross-species and zero-shot transfer.

Improving nanobody structure prediction with self-distillation

This study improves nanobody structure prediction by using synthetic data and a classifier to filter structural accuracy, achieving better results than existing methods with lower computational cost.

Cortex-Wide Preservation of Multi-Stimulus Information across Degrees of Network Synchronization

The paper decodes stimulus identity and network state from calcium imaging in mice across different anesthesia states, showing preserved stimulus information even during deep anesthesia.

A Comprehensive Evaluation of Self Attention for Detecting Regulatory Feature Interactions

This study evaluates attention-based methods for detecting transcription factor cooperativity, showing that entropy-enhanced models improve interpretability and precision.

Improving cell-type identification with generative adversarial networks-enhanced augmentation-free single-cell RNA-Seq contrastive learning

This paper introduces GAN-RCL, a contrastive learning method for cell-type identification in single-cell RNA-Seq that outperforms existing methods.

multiVIB: A unified probabilistic contrastive learning framework for atlas-scale integration of single-cell multi-omics data

multiVIB integrates single-cell multi-omics data across modalities and species with a probabilistic contrastive learning framework, achieving state-of-the-art performance while preserving biological variation.

DeepCor: Denoising fMRI Data with Contrastive Autoencoders

DeepCor improves fMRI denoising by using contrastive autoencoders, outperforming CompCor on simulated data and enhancing connectivity differences in real datasets.

Rewriting protein alphabets with language models

This paper introduces a new 20-letter alphabet, TEA, derived from protein language models, enabling efficient homology searches comparable to structure-based methods without structural data.

Rapid and accurate protein structure database search using inverse folding model and contrastive learning

The paper introduces mTM-align2, a method for rapid and accurate protein structure database search that uses inverse folding models and contrastive learning, achieving high precision in seconds.

LaCONIC: A Label-Aware and Graph-Guided Contrastive Multi-Omics Collaborative Learning Model for Cancer Risk Prediction

LaCONIC improves cancer survival prediction by integrating multi-omics data with gene regulatory networks and cancer subtypes, achieving better performance and interpretability.

SIRP: A Self-Supervised Fluorescence Denoising method with Implicit Representation Priors

SIRP improves fluorescence microscopy image denoising by using self-supervised learning and implicit neural representations to preserve fine details under low SNR conditions.

RamanMAE: Masked Autoencoders Enable Efficient Molecular Imaging by Learning Biologically Meaningful Spectral Representations

RamanMAE uses masked autoencoders to learn spectral representations from Raman data, enabling efficient processing and noise reduction in biological imaging.

iDRKAN: Interpretable miRNA-Disease Association Prediction Based on Dual-Graph Representation Learning and Kolmogorov-Arnold Network

iDRKAN predicts miRNA-disease associations using dual-graph representation learning and an interpretable Kolmogorov-Arnold Network, outperforming existing methods on public datasets.

REECAP: Contrastive learning of retinal aging reveals genetic loci linking morphology to eye disease

REECAP uses contrastive learning to link retinal aging patterns with genetic loci associated with eye diseases, identifying 178 loci including 27 linked to age-related eye diseases.

Evo2HiC: a multimodal foundation model for integrative analysis of genome sequence and architecture

Evo2HiC improves prediction of 3D genome architecture and epigenomic signals by integrating DNA sequences and Hi-C data, achieving better accuracy and computational efficiency.

Contrastive modelling of transcription and transcript abundance in legumes using PlanTT

The paper builds contrastive models to predict expression differences between ortholog pairs using a multiomic dataset from legumes.