contrastive learning

Cross-Propagative Graph Learning Reveals Spatial Tissue Domains in Multi-Modal Spatial Transcriptomics

This paper introduces st-Xprop, a graph learning method for spatial domain detection in multi-modal spatial transcriptomics, improving clustering accuracy and biological relevance.

Metagenomic contextualization of proteins with state space models

This paper introduces Nammu, a state-space model for metagenomic data that outperforms Transformer-based models in taxonomic and functional predictions.

Conformation-Dependent Donor Selectivity in the Xanthan Gum Glycosyltransferase GumK Revealed by AI-Based Docking

This study uses AI-based docking and cofolding to show that GumK's donor selectivity depends on conformational states and ligand interactions, not a single rigid binding mode.

Cross-architecture ensembling of DNA foundation models improves the precision and stability of chimera detection in long-read metagenomic bins

The paper evaluates DNA foundation models for chimera detection in long-read metagenomic bins, finding that ensembling improves precision over single models.

Why linkage disequilibrium measures disagree: Fisher geometry of rare common haplotype structure

The paper analyzes why LD measures disagree in the rare-common regime by examining the geometric structure of the haplotype simplex under the Fisher information metric.

HetNetEX: Exact Asymptotic Inference in Heterogeneous Biomedical Knowledge Graphs

HetNetEX provides exact asymptotic inference for connectivity in heterogeneous biomedical knowledge graphs, outperforming XSwap in speed and accuracy.

Tissue tearing degrades optimal-transport and diffeomorphic registration of spatial transcriptomics beyond displacement magnitude: a multi-seed deformation benchmark and a supervised graph cross-attention proof-of-concept.

This paper benchmarks registration methods for spatial transcriptomics under tissue tearing and introduces a supervised graph model, finding that tearing degrades performance across methods and that a learned model performs better in-sample but not across donors.

Connectome-scale self-supervised representation learning reveals neuronal organization beyond canonical labels

This paper develops a self-supervised framework for learning structure-connectivity representations in dense connectomes, revealing neuronal organization beyond canonical labels.

How Not to be Seen: Predicting Unseen Enzyme Functions using Contrastive Learning

This paper introduces EnzPlacer, a contrastive learning method that predicts enzyme function from sequence by placing it in known function space, even when the precise function is unknown.

A PERTURBATION CELL ATLAS OF HUMAN INDUCED PLURIPOTENT STEM CELLS

The paper generates a CRISPRi perturbation cell atlas in human induced pluripotent stem cells, identifies novel regulators of pluripotency and metabolism, and proposes a machine learning framework for predicting perturbation recipes.

NeuroVLM: A generative vision-language framework for human neuroimaging

NeuroVLM generates and evaluates vision-language models for human neuroimaging using 30,826 neuroimage-text pairs to perform tasks like text-to-image generation and image-to-text interpretation.

CPS: Mapping Physical Coordinates to High-Fidelity Spatial Transcriptomics via Privileged Multi-Scale Context Distillation

CPS maps physical coordinates to high-fidelity spatial transcriptomics using a multi-scale context distillation strategy, achieving state-of-the-art imputation and denoising performance.

Fairness-aware Supervised Hierarchical Contrastive Semantic Learning for Sexual Dimorphism Analysis

This paper introduces FairHICON, a fairness-aware supervised hierarchical contrastive learning method that identifies sex-specific and common genomic drivers, improving predictive performance and reducing sex-based performance gaps in cancer and asthma datasets.

Graph-based Contrastive Learning Enables Unified Integration and Niche Transfer Across Single-Cell and Spatial Multi-Omics

Garfield integrates single-cell and spatial multi-omics data into a shared space, outperforming existing methods in identifying spatial niches and revealing immune microenvironment heterogeneity.

USP-ddG: A Unified Structural Paradigm with Data Efficacy and Mixture-of-Experts for Predicting Mutational Effects on Protein-Protein Interactions

USP-ddG predicts mutational effects on protein-protein interactions by improving data efficiency and modeling conformational dynamics, outperforming existing methods on benchmark datasets.

A hyperbolic topological atlas reveals polyamine steering of a shared developmental manifold in Arabidopsis

The paper uses a hyperbolic topological atlas to analyze Arabidopsis rosette development under different polyamine and nutrient conditions, revealing how these factors influence developmental trajectories.

Self-Supervised Behavioral Representations Across the Life Course: A Killifish Case Study

This paper develops a self-supervised model to predict aging and lifespan from behavioral data in killifish, finding that single-day predictions can match whole-life models for some traits.

Intra-African Geographic Domain Shift in Wildlife Camera Trap Species Classification: A Comparative Study of Supervised and Zero-Shot Foundation Models

This study evaluates how well wildlife camera trap species classification models perform across different African regions, comparing supervised and zero-shot models.

Learning Perturbation Effects Through Contrastive Alignment of Multimodal Biological Embeddings

This paper introduces PertOmni, a multimodal framework that aligns transcriptomic data with text and image embeddings to improve perturbation effect prediction.

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.

Zero-shot design of drug-binding proteins via neural selection-expansion

This paper introduces a method using two neural networks to design drug-binding proteins with high affinity, surpassing current state-of-the-art methods.

Deep Transfer Learning for Dormancy and Outbreaking State Classification in Metastatic Breast Tumor Cells: A Benchmark of Modern Deep Learning Models

This study benchmarks deep learning models for classifying dormant versus proliferative breast tumor cells using spatiotemporal data from a 3D BME system, finding EfficientNet-B7 achieves high accuracy.

Self-Calibrated Hyperspectral Neural Radiance Fields for 3D Reconstruction of Bone and Bone Analogues

This paper introduces BoNeRF-HS, a self-calibrated hyperspectral neural radiance field for 3D reconstruction of bone and bone analogues, achieving better reconstruction quality and surface detail preservation.

MISO: A Controlled Ablation of Masking, Initialization, Sampling, and Optimization for Segmentation in Volumetric Electron Microscopy

The paper evaluates sampling, loss functions, and masking strategies for vEM organelle segmentation, finding that these factors significantly impact performance, especially for rare classes.

πDIA-CLIP: efficient identification of highly heterogeneous proteomics data via a generalized zero-shot framework

The paper introduces πDIA-CLIP, a zero-shot framework for DIA proteomics data analysis that improves protein identification and reduces false positives across diverse datasets.

Nevermore: Target-Conditioned Protein-Ligand Representation Learning for Multi-Objective Lead Optimization with Database-Grounded Retrieval

Nevermore improves lead optimization by balancing binding affinity and developability constraints using a database-grounded framework with interpretable molecular edits.

Microscopy-informed structural connectivity mapping in the in vivohuman brain via domain adaptation

This paper develops a deep learning model using domain adaptation to predict high-resolution fibre orientations from diffusion MRI, informed by microscopy data, improving connectivity mapping in human brains.

Linear-time prediction of proteome-scale microbial protein interactions

FlashPPI predicts proteome-scale microbial protein interactions in linear time using contrastive learning and genomic language models, outperforming existing methods and enabling rapid network analysis.

DOMINO: graph diffusion learning identifies spatial domain structures with enhanced accuracy and scalability

DOMINO identifies spatial domains in spatial transcriptomic data with improved accuracy and scalability, outperforming existing methods on large datasets.

Linking automated image analysis to ecological inference: high-throughput monitoring of soil fauna

This paper develops an efficient image processing pipeline for soil fauna monitoring by combining traditional computer vision with fine-tuned foundation models, reducing annotation effort and enabling large-scale ecological analysis.