contrastive learning

BiCLUM: Bilateral Contrastive Learning for Unpaired Single-Cell Multi-Omics Integration

BiCLUM integrates unpaired single-cell multi-omics data by aligning cell and feature spaces through bilateral contrastive learning, outperforming existing methods in visualization and quantitative metrics.

Geometric Fragility in Alzheimer's Disease: Probing the Loss of Hippocampal Hierarchical Abstraction via Contrastive Point Cloud Modeling

This study uses contrastive point cloud modeling to show that Alzheimer's disease involves a loss of hippocampal hierarchical abstraction, revealing a linear geometric degradation correlated with cognitive decline.

MolSpecFlow: Mass-Constrained Hybrid Flow Matching for Joint Molecular-Spectral Analysis

MolSpecFlow bridges the gap between molecules and mass spectra using a hybrid flow matching model, achieving state-of-the-art results in de novo generation and spectral simulation.

An Artificial Intelligence-based framework for protein interaction design with accelerated KAN-based Positive-Unlabeled learning

The paper introduces an AI framework for protein interaction design using KAN-based PU learning, achieving improved stability and validation metrics compared to traditional methods.

FLARE: Fine-grained Learning for Alignment of spectra-molecule REpresentation Enhances Metabolite Annotation

FLARE improves metabolite annotation by aligning spectra and molecular structures through contrastive learning, achieving state-of-the-art performance on MassSpecGym.

stGCL: A versatile cross-modality fusion method based on multi-modal graph contrastive learning for spatial transcriptomics

stGCL integrates multi-modal spatial transcriptomics data using graph contrastive learning to improve clustering and spatial pattern analysis.

scAURA: Alignment- and Uniformity-based Graph DebiasedContrastive Representation Architecture for Self-SupervisedClustering of Single-Cell Transcriptomics

scAURA improves cell type clustering in scRNA-seq by integrating graph debiased contrastive learning, outperforming existing methods across multiple datasets.

FoldVision: A compute-efficient atom-level 3D protein encoder

FoldVision is a compute-efficient 3D CNN for proteins that outperforms sequence-based models on structure-dependent tasks using atom-level representations.

High-fidelity bioimage restoration via adversarial learning

This paper introduces NAFNet GAN, a bioimage restoration framework that achieves high fidelity and low latency, suitable for real-time microscopy applications.

Conformalized Multiview Learning

This paper introduces Coracle, a conformalized framework for multimodal AI that provides valid uncertainty estimates, outperforming existing methods in uncertainty estimation and calibration.

Fast Adversarial Generation of Molecular Dynamics Trajectories with Kinetic Fidelity

This paper introduces a GAN-based method to generate molecular dynamics trajectories with kinetic fidelity, enabling efficient sampling of rare events in complex systems.

SigFormer: an Attention-Based Framework for Robust Single-Sample Mutational Signature Decomposition

SigFormer improves single-sample mutational signature decomposition by using a transformer framework, outperforming existing methods in high-noise and overcomplete settings.

Exploring phenotype-related single-cells through attention-enhanced representation learning

This study introduces a method to predict clinical phenotypes from single-cell data using attention-based multiple instance learning, demonstrating its effectiveness in predicting disease severity and age-related phenotypes.

DNACSE: Enhancing Genomic LLMs with Contrastive Learning for DNA Barcode Identification

DNACSE improves DNA barcode identification by fine-tuning DNA language models with contrastive learning, achieving higher accuracy and clustering performance than existing methods.

COMIC: Explainable Drug Repurposing via Contrastive Masking for Interpretable Connections

COMIC improves drug repurposing prediction with explainable pathways, achieving 9.55% better performance than existing methods on a large knowledge graph.

Hierarchical bounds on RNA chromatin statistical dependence across cellular states in paired single-cell multiome data

The paper quantifies RNA-chromatin statistical dependence across cellular states using paired single-cell RNA-seq and ATAC-seq data, finding that most dependence is due to cell-type composition rather than regulatory coordination.

Deep contrastive learning enables genome-wide virtual screening

DrugCLIP, a contrastive dense-retrieval model for pockets and molecules, beats docking on DUD-E and LIT-PCBA and screened about 10,000 human AlphaFold2 proteins in 24 hours.

Computational design of conformation-biasing mutations to alter protein functions

Contrastive scoring of protein variants across alternative backbone structures with inverse folding models (Conformational Biasing) predicts mutations that shift conformational equilibria and tune effector binding, catalysis, and labeling specificity.

Cross-Domain Transfer Learning from Peptides to Lipids Using a Multi-Property Fine-Tuned LLM

The paper uses a multi-property fine-tuned language model to predict lipid retention times from peptide data, improving accuracy in data-sparse scenarios.

Uncertainty-aware genomic deep learning with knowledge distillation

This paper introduces DEGU, a method that improves deep learning reliability in genomics by combining ensemble learning and knowledge distillation to capture both average predictions and uncertainty.

PRADA-DTI: A Prototype-Retrieval Augmented Domain-Adaptation Framework for Drug-Target Interaction Prediction

PRADA-DTI improves drug-target interaction prediction by enabling continual learning from new protein domains with minimal parameter updates and reduced forgetting.

Revealing a coherent cell state landscape across single cell datasets with CONCORD

CONCORD is a self-supervised framework that integrates single-cell datasets across batches, technologies, and species, producing high-resolution cell atlases with improved biological resolution and reduced noise.

HistoSB-Net: Semantic Bridging for Data-Limited Cross-Modal Histopathological Diagnosis

HistoSB-Net adapts pre-trained vision-language models for histopathological diagnosis with limited data, improving performance through semantic bridging.

Majority-LCL: Towards Malaria Cell Detection using Label Contrastive Learning and Majority voting Ensembling

This paper introduces a malaria cell detection method combining label contrastive learning and majority voting ensembling to improve diagnostic accuracy using microscopic images.

NeuroConText: Contrastive Learning for Neuroscience Meta-Analysis with Rich Text Representation

NeuroConText improves brain meta-analysis by using contrastive learning and large language models to better retrieve and reconstruct brain activation maps from neuroscience texts.

Cell segmentation without annotation by unsupervised domain adaptation based on cooperative self-learning

This paper introduces an unsupervised domain adaptation method for cell segmentation that avoids predefined thresholds, achieving accuracy comparable to supervised models without labeled data.

Scalable embedding fusion with protein language models: insights from benchmarking text-integrated representations

The paper benchmarks text-integrated protein language models and introduces embedding fusion to improve performance across diverse tasks.

Frustratingly easy domain adaptation for cross-species transcription factor binding prediction

The paper introduces MORALE, a domain adaptation framework for cross-species TF binding prediction, achieving state-of-the-art performance without adversarial training.

Holotomography-driven learning for in-silico staining of single cells in flow cytometry avoiding co-registration

This paper introduces a Holotomography-driven learning method for virtual staining in flow cytometry without requiring co-registration, enabling label-free single-cell analysis.

Calibration of MRI-based reference intervals to new samples

This paper introduces ReForm, a method to calibrate MRI-based reference intervals for new samples without sharing reference data, showing comparable or better false positive rate control than alternative methods.