contrastive learning

Evaluation of Transfer Learning Methods for Detecting Alzheimer's Disease with Brain MRI

This paper evaluates transfer learning methods for Alzheimer's disease classification using brain MRI, finding that contrastive learning improves performance, especially with limited data.

Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST

GraphST improves spatial transcriptomics analysis by integrating clustering, integration, and deconvolution through graph self-supervised contrastive learning, achieving higher accuracy than existing methods.

EiDA: A lossless approach for the dynamic analysis of connectivity patterns in signals; application to resting state fMRI of a model of ageing

EiDA is a method for analyzing dynamic functional connectivity in resting-state fMRI data from ageing rats, revealing reduced phase-locking intensity and structural complexity with age.

High-fidelity Database-free Deep Learning Reconstruction for Real-time Cine Cardiac MRI

This paper introduces a zero-shot self-supervised physics-guided deep learning method for real-time cine cardiac MRI, showing improved performance over existing techniques in noise and aliasing reduction.

Dissecting cell state transitions by causal inference

CIBER is a causal inference framework for analyzing cell state transitions, identifying key features in differentiation processes and predicting novel regulatory associations.

GeneDART: Extending gene coverage in image-based spatial transcriptomics by deep learning-based domain adaptation with barcode-based RNA-sequencing data

GeneDART extends gene coverage in imaging-based spatial transcriptomics by combining it with barcode-based data using deep learning for domain adaptation.

scAEGAN: Unification of Single-Cell Genomics Data by Adversarial Learning of Latent Space Correspondences

scAEGAN unifies single-cell genomics data by adversarial learning of latent space correspondences, outperforming existing methods in integration and prediction tasks.

The role of the Frontal Aslant tract in bilingual language control

This study examines the role of the Frontal Aslant tract in bilingual language control by comparing unimodal and bimodal bilinguals, finding distinct associations with language production tasks.

Characterising stationary and dynamic effective connectivity changes in the motor network during and after tDCS

This study investigates how tDCS affects effective connectivity in the motor network, finding widespread effects beyond the target area with unique temporal patterns.

Protein Representation Learning via Knowledge Enhanced Primary Structure Modeling

The paper introduces KeAP, a knowledge-enhanced auto-encoder for protein representation learning that outperforms existing methods on multiple downstream tasks.

scTriangulate, a game-theory based framework for optimal solutions of uni- and multimodal single-cell data

scTriangulate identifies high-confidence cell populations and modality-specific subtypes by combining clustering results from uni- and multimodal single-cell data using a game-theory framework.

Active Learning of the Conformational Ensemble of Proteins using Maximum Entropy VAMPNets

This paper introduces MaxEnt VAMPNet, an active learning method that accelerates conformational landscape exploration in biomolecules by maximizing entropy in VAMPNet-trained microstates.

Phenotypic subtyping via contrastive learning

This paper introduces PACA, a contrastive learning method for subphenotype subtyping, which outperforms existing methods by capturing weak subphenotypic variation in complex diseases.

Single-cell Multi-omics Integration for Unpaired Data by a Siamese Network with Graph-based Contrastive Loss

This paper introduces MinNet, a Siamese network with graph-based contrastive loss for integrating single-cell multi-omics data, showing its effectiveness in removing batch effects and improving downstream analysis.

Energy Guided Diffusion for Generating Neurally Exciting Images

This paper introduces Energy Guided Diffusion (EGG) to generate neurally exciting images, outperforming gradient ascent in generalization and efficiency for macaque V4 models.

Deep Learning Based Registration of Serial Whole-slide Histopathology Images in Different Stains

This paper introduces CycGANRegNet, a deep learning method for registering serial WSIs in different stains, achieving better performance than existing methods on breast cancer patient data.

scTopoGAN: Unsupervised manifold alignment of single-cell data

scTopoGAN aligns single-cell datasets with non-overlapping cells or features using topology-guided GANs, outperforming existing methods in unsupervised settings.

FLONE: fully Lorentz network embedding for inferring novel drug targets

FLONE improves drug target prediction by using hyperbolic space embeddings to better capture hierarchical relationships in drug-disease-target networks.

Contrast Subgraphs Allow Comparing Homogeneous and Heterogeneous Networks Derived from Omics Data

The paper introduces contrast subgraphs to compare gene and protein networks from different conditions or techniques, highlighting altered connectivity modules.

A 3D Clinical Face Phenotype Space of Genetic Syndromes using a Triplet-Based Singular Geometric Autoencoder

This paper develops a triplet-based geometric autoencoder to create a clinical face phenotypic space for syndrome diagnosis, showing it can classify syndromes, generalize to new ones, and preserve genetic disease relationships.

Enhanced compound-protein binding affinity prediction by representing protein multimodal information via a coevolutionary strategy

This paper introduces FeatNN, a coevolutionary strategy for predicting compound-protein binding affinity by jointly representing protein sequence and structure information, achieving higher accuracy than existing methods.

3D Graph Contrastive Learning for Molecular Property Prediction

This paper introduces a small-scale 3D Graph Contrastive Learning framework for molecular property prediction, achieving state-of-the-art results with limited resources and 3D structural information.

Graph neural networks learn emergent tissue properties from spatial molecular profiles

This paper evaluates graph neural networks' ability to predict tissue-level phenotypes from spatial molecular data, finding that some phenotypes can be captured without spatial information and that overfitting limits representation learning.

CoAtGIN: Marrying Convolution and Attention for Graph-based Molecule Property Prediction

The paper introduces CoAtGIN, a graph neural network combining convolution and attention for molecule property prediction, achieving high performance with fewer parameters.

Image-based quantification of Arabidopsis thaliana stomatal aperture from leaf images

This paper presents an automated image analysis pipeline for quantifying stomatal aperture in Arabidopsis thaliana leaves using brightfield microscopy images.

Biologically plausible unsupervised learning in neural networks with sparse and asymmetric connectivity

The paper develops biologically plausible variants of RBMs with sparse and asymmetric connectivity and benchmarks their performance on MNIST.

Spatial scale influences taxon conservation in the eukaryotic virome of a mosquito

The paper examines how spatial scale affects taxon conservation in the eukaryotic virome of Culex pipiens mosquitoes across multiple countries, revealing deterministic patterns in virome assembly.

OPUS-DSD: Deep Structural Disentanglement for cryo-EM Single Particle Analysis

OPUS-DSD improves cryo-EM reconstruction by resolving structural heterogeneity and enhancing homogeneity in datasets, outperforming existing methods on synthetic and real data.

Assessment of Emerging Pretraining Strategies in Interpretable Multimodal Deep Learning for Cancer Prognostication

This study evaluates pretraining strategies in an interpretable multimodal deep learning framework for cancer prognosis, showing that pretraining improves performance over standard methods.

scFormer: A Universal Representation Learning Approach for Single-Cell Data Using Transformers

scFormer uses transformers to learn cell and gene embeddings for single-cell data, achieving state-of-the-art results on multiple tasks.