contrastive learning

Language-guided domain adaptation improves protein sequence rep-resentations for antimicrobial resistance analysis

BacLLM improves protein sequence representations for antimicrobial resistance analysis by integrating functional knowledge and enhancing resistance gene identification and mechanism prediction.

Reference-Guided De Novo Retrieval of 3D Chromatin Organizational Features with Representation Learning

The paper introduces C-Fetch, a framework that retrieves 3D chromatin features using reference examples, showing improved enrichment for biologically relevant contacts compared to existing methods.

Perturbation-induced gene programs map the hidden regulatory architecture of complex diseases

This paper introduces Perturbscape, a framework that uses perturbation-induced gene programs to identify disease-relevant genes and pathways, showing higher heritability informativeness than steady-state approaches.

RELAY: Relational In-Context Learning for Cellular Perturbation Prediction

RELAY predicts cellular responses to perturbations by learning relational patterns from measured data, outperforming existing methods on unseen variable values and combinations.

FedEpitope: Parameter-Efficient Federated Protein Language Models for Linear B-Cell Epitope Prediction

FedEpitope uses federated learning with LoRA to predict linear B-cell epitopes without centralizing data, achieving good performance across multiple clients.

MaxCAT-Net: Hybrid Transformer-Based Skin Lesion Classification and Lightweight Segmentation with Calibration and Cross-Dataset Validation

MaxCAT-Net improves skin lesion classification and segmentation with hybrid architecture and calibration, showing performance on benchmark datasets but with preliminary results.

A single-cell multimodal foundation model enables cross modality analysis between transcriptome and chromatin

MUTUAL enables cross-modality analysis between transcriptome and chromatin by learning a shared RNA-ATAC space from multiome data without refitting.

Neopep: Reinforcement Learning From Machine-Learned Force Field Energy for Non-Canonical Peptide Structure Prediction

Neopep uses reinforcement learning with machine-learned force field energy to predict structures of non-canonical amino acid-containing peptides, achieving lower RMSD than existing methods.

VenusGenome: Genome-Aware Masked Autoencoding for Microbial Genome Representation Learning

VenusGenome improves microbial genome representation learning by using a genome-aware masked autoencoder, achieving high performance in phenotype prediction and gene-level tasks.

Improved cryo-EM reconstruction of sub-50 kDa complexes using 2D template matching

This paper improves cryo-EM reconstruction of small complexes using 2D template matching and high-resolution structures as priors, enhancing density of ligand-binding sites in a 43 kDa protein kinase.

LoCoFlow: Local-Context Flow Matching for Single-Cell Dynamics and Fate Inference

LoCoFlow improves trajectory inference in single-cell data by incorporating local geometry and fate uncertainty, achieving better performance than existing methods.

Systematic comparison of color representations between humans and deep neural networks: towards predicting human color perception in a vast color space

This study compares how different learning paradigms enable deep neural networks to predict human color perception across a vast color space.

REMAG: recovery of eukaryotic genomes from metagenomic data using non-contrastive self-supervised learning

REMAG improves eukaryotic genome recovery from metagenomic data by using non-contrastive self-supervised learning, outperforming existing tools in producing near-complete genomes.

From Sparse Visits to Continuous Molecular Trajectories with Flow Matching

CohortFM reconstructs continuous molecular trajectories from sparse longitudinal data, revealing biologically meaningful progression patterns across diseases.

CRISCross: Integration of Epigenetic Information for Deep-Learning-Based Prediction of CRISPR-Cas9 Off-Targets Across Cell Types

CRISCross improves CRISPR-Cas9 off-target prediction by integrating epigenetic features with sequence data across cell types, showing better performance than sequence-only models.

UniMedSeg: Unified Multi-Modal Medical Image Segmentation via Uncertainty Guided Encoding and Frequency-Spatial Dual-Prompt Decoding

UniMedSeg improves multi-modal medical image segmentation by integrating uncertainty-guided encoding and dual-prompt decoding for better domain generalization and spatial risk estimation.

Frustration Quenching and Network Topology of the Energy Landscape as Primary Determinants of Protein-Ligand Binding Pose Prediction by Deep Learning Models

This paper shows that deep learning models' performance in predicting protein-ligand binding poses depends on the energy landscape's frustration quenching and network topology, not algorithmic flaws.

MINT: Predicting Enzyme-Cofactor Specificity by Multimodal Knowledge Transfer from Protein-Ligand Interactions

MINT predicts enzyme-cofactor specificity using a multimodal framework with transfer learning, achieving high accuracy in cofactor binding prediction.

Cross-cohort Mamography]{Representation-dependent domain adaptation for cross-cohort generalization of mammography diagnostic models

This study evaluates cross-cohort generalization of mammography diagnostic models using different image representations and domain-adversarial learning, finding that MedSigLIP embeddings provide the best transferability.

trFlow: ultrafast all-atom protein conformational ensemble generation via conditional flow matching

trFlow generates all-atom protein conformations faster and more accurately than existing methods by using flow matching with three key innovations.

Deep learning of fossil pollen morphology reveals 25,000 years of ecological change in East African grasslands

This paper uses deep learning to analyze fossil grass pollen morphology and reconstruct 25,000 years of ecological change in East African grasslands, showing reduced diversity and shifts in C3:C4 ratios during the last ice age.

Multi-Model Machine Learning Consensus Identifies a Dual-Interferon Gene Signature in Ischemic Stroke: An Integrative Single-Cell Transcriptomic Analysis

This paper uses a multi-model consensus framework to identify a dual-interferon gene signature in ischemic stroke using single-cell RNA sequencing data from a mouse model.

CycloCross: Adversarial single-cell RNA-seq data translation across species

Translating cells between mouse, macaque and human scRNA-seq with a CycleGAN, without ortholog mapping, preserved developmental time structure, imputed missing states and produced more realistic synthetic cells than scGAN.

RNASeek: A Cross-Phyla Generative Foundation Model for Multipurpose RNA Modeling and Reinforcement Learning-Based Design

RNASeek is a generative foundation model for RNA sequence design and functional prediction, using reinforcement learning to optimize RNA properties like stability and cleavage activity.

Tri-Modality Representation Learning for Molecular Property Prediction

This paper introduces TMCA, a multimodal framework that integrates SMILES, molecular graphs, and 3D conformations for molecular property prediction, achieving the best performance on benchmark datasets.

A Taxonomy-Informed Sparse DNA Foundation Model for Microbial Genomics

A sparse mixture-of-experts DNA model conditioned on hyperbolic taxonomy embeddings learns better microbial representations than its taxonomy-free twin and comes close to Evo2-7B in probing and fine-tuning at lower compute.

Advancing Wildlife Image Analysis: A Graph Attention Contrastive Learning Approach for Region-Specific Mammal Classification

This paper develops a region-specific AI model for mammal classification using graph attention contrastive learning, achieving high accuracy in recognizing local species and improving data cleaning efficiency.

AMADEUS: Annotation-free multi-animal direction estimation using self-supervised learning

AMADEUS tracks multiple animals without annotations or markings, achieving better performance in identity tracking and keypoint localization across various species.

SHOT-CCR: Biologically guided adversarial training for test-time adaptation in cellular morphology

SHOT-CCR improves classification of genetic perturbations in Cell Painting datasets by reducing batch effects through biologically guided adversarial training.

Scalable Computational Phenomics of Nuclear Morphology

NucXplore converts H&E-stained nuclei into 129 explicit features, outperforming classical and deep learning methods in histopathology cohorts and revealing aging and exposure-related cellular changes.