contrastive learning

Icecream: High-Fidelity Equivariant Cryo-Electron Tomography

Icecream improves cryo-ET reconstruction by addressing denoising, missing-wedge recovery, and computational efficiency using equivariant deep learning.

Odyssey: reconstructing evolution through emergent consensus in the global proteome

Odyssey generates and designs proteins using a multimodal language model with a consensus mechanism, achieving strong performance in protein generation and structure discretization.

ECloudGen: Leveraging Electron Clouds as a Latent Variable to Scale Up Structure-based Molecular Design

ECloudGen improves molecular design by using electron clouds as latent variables, generating more potent binders and exploring broader chemical space than existing methods.

scConcept: Contrastive pretraining for technology-agnostic single-cell representations beyond reconstruction

scConcept improves single-cell representations by using contrastive learning instead of reconstruction, outperforming existing methods in multiple tasks.

Mapping the diverse topologies of protein-protein interaction fitness landscapes

The paper identifies target-specific fitness landscapes in protein-protein interactions and develops a supervised contrastive learning approach to predict binders.

Integrative machine learning predicts activating kinase mutations for precision oncology

Kinome-AI predicts activating kinase mutations using machine learning, achieving high accuracy without requiring structural inputs for new mutations.

Multiple instance learning with spatial transcriptomics for interpretable patient-level predictions: application in glioblastoma

This paper introduces SpaMIL, a spatial transcriptomics-based multiple instance learning framework for patient-level outcome prediction in glioblastoma, achieving high prognostic accuracy.

Age-informed, attention-based weakly supervised learning for neuropathological image assessment

This paper introduces an age-informed attention-based MIL model to predict AT8 density in neuropathological images, improving diagnostic accuracy and interpretability.

Nanoscale spatial-omics via contrastive embedding of single-molecule localisation data

This paper introduces a contrastive learning framework for nanoscale spatial-omics, enabling analysis of single-molecule localisation data to study protein architecture and its changes under various conditions.

Learning universal knowledge graph embedding for predicting biomedical pairwise interactions

This paper introduces LukePi, a self-supervised pre-training framework for GNNs on biomedical knowledge graphs, which improves link prediction in low-data and distribution-shift scenarios.

Understanding Virtual Staining with generative adversarial networks for Osteoclast Imaging

This study uses generative adversarial networks to perform virtual staining of osteoclast cultures, showing that input image background influences staining predictions and enabling non-invasive analysis.

Representations in the hippocampal-entorhinal system emerge from learning sensory predictions

This paper introduces a computational model that generates diverse neural representations in the hippocampal-entorhinal system through contrastive learning, capturing spatial coding patterns observed in real neural data.

Close-Up of vesicular ER Exit Sites by Volume Electron Imaging using FIB-SEM

The paper uses FIB-SEM and ASEM to identify and characterize vesicular ER exit sites in mammalian cells, revealing their abundance and organization.

ProTrek: Navigating the Protein Universe through Tri-Modal Contrastive Learning

ProTrek introduces a tri-modal contrastive learning model for protein sequence, structure, and function, achieving significant improvements in speed and accuracy over existing tools.

MSACLR: Contrastive Learning of Protein Conformations from MSAs

MSACLR uses contrastive learning to map multiple sequence alignments to protein conformations, achieving better fold separation and capturing conformational variation.

Peptide Design through Binding Interface Mimicry

PepMimic, an all-atom latent diffusion model guided by a contrastive interface encoder, designs peptides mimicking known binders; 8% of tested peptides across five targets reached 10^-8M KD.

Isolation of Pure Disease Specific Aging Trajectories in Spatial Transcriptomics via the Delta Delta Method

The paper introduces the Delta-Delta Method to isolate disease-specific aging trajectories in spatial transcriptomics by subtracting wild-type aging from transgenic trajectories, revealing distinct molecular programs in neurodegeneration.

InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders

This paper introduces InterPLM, a framework using sparse autoencoders to extract interpretable features from protein language models, revealing both known and novel biological concepts.

scPhoenix: A Contrastive Learning-based Framework with Aux-Core Feature Disentanglement Enhances Sparse Cellular Multi-omics Translation

scPhoenix enhances sparse cellular multi-omics translation using contrastive learning and aux-core feature disentanglement, preserving heterogeneity and improving performance in high sparsity and spatial heterogeneity scenarios.

Prediction of fast subcellular calcium signals based on label-free second harmonic generation microscopy measurements

The paper predicts calcium sparks using label-free SHG microscopy and pix2pix, showing agreement with ground truth data on frequency and location.

High-Fidelity Neural Speech Reconstruction through an Efficient Acoustic-Linguistic Dual-Pathway Framework

This paper introduces a dual-path framework for reconstructing high-fidelity speech from neural recordings, achieving high intelligibility with limited ECoG data.

Steering Vector Fields for Property-Controlled Molecular Generation with Chemical Language Models

This paper improves molecular generation by steering chemical language models to control properties like solubility and lipophilicity using activation steering techniques.

Small Molecule Approach to RNA Targeting Binder Discovery (SMARTBind) Using Deep Learning Without Structural Input

SMARTBind uses deep learning and contrastive learning to identify small molecule binders to RNA without structural input, outperforming existing methods and reducing computational cost.

Exploring Augmentation-Driven Invariances for Graph Self-supervised Learning in Spatial Omics

This study evaluates how different graph augmentations affect embedding quality and downstream performance in spatial omics, showing that well-designed augmentations improve results while poor ones can harm performance.

Domain-Invariant Feature Learning for Patient-Level Phenotype Prediction from Single-Cell Data

The paper improves patient-level phenotype prediction from scRNA-seq data by learning domain-invariant features, showing better out-of-domain accuracy and biological fidelity.

PI-FC: Pre-training Individual-specific Functional Connectome through State-invariant Contrastive Learning

PI-FC uses state-invariant contrastive learning to extract stable individual brain signatures from fMRI, improving phenotypic prediction accuracy and reducing scanning time across multiple datasets.

Learning Residual-based Biomarkers of Cognitive Health via Self-Supervised Learning on EEG State Transitions

This paper introduces a self-supervised learning framework for EEG-based biomarker discovery of cognitive health, but finds poor predictive performance using plasma p-tau217 as a proxy.

An extensive evaluation of single-cell RNA-Seq contrastivelearning generative networks for intrinsic cell-typesdistribution estimation

This study evaluates contrastive learning generative networks for estimating cell-type distributions in single-cell RNA-Seq data, finding supervised methods outperform self-supervised ones.

High-throughput virus quantification using cytopathic effect area analysis by deep learning

This paper develops a deep learning-based method to quantify infectious viruses using cytopathic effect area analysis, improving throughput and reducing manual bias.

Unify: Learning Cellular Evolution with Universal Multimodal Embeddings

Unify learns universal cell embeddings across species by integrating RNA expression with protein and general language model embeddings, enabling cross-species comparisons and accurate perturbation response prediction.