PathOmCLIP: Connecting tumor histology with spatial gene expression via locally enhanced contrastive learning of Pathology and Single-cell foundation model DOI Creative Commons
Yong‐Ju Lee, Xinhao Liu, Minsheng Hao

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 11, 2024

Abstract Tumor morphological features from histology images are a cornerstone of clinical pathology, diagnostic biomarkers, and basic cancer biology research. Spatial transcriptomics, which provides spatially resolved gene expression profiles overlaid on images, offers unique opportunity to integrate features, thereby deepening our understanding tumor biology. However, spatial transcriptomics experiments with patient samples in either trials or care costly challenging, whereas generated routinely available for many legacy prospective cohorts disease progression outcomes well-annotated cohorts. Inferring computationally these would significantly expand biology, but paired data training multi-modal spatial-histology models remains limited. Here, we tackle this challenge by incorporating performant foundation pre-trained massive datasets pathology single-cell RNA-Seq, respectively, provide useful embeddings underpin models. To end, developed PathOmCLIP, model trained contrastive loss create joint-embedding space between histopathology RNA-seq model. We incorporate set transformer gather localized neighborhood architecture following training, further enhances performance is necessary obtain robust results. validate PathOmCLIP across five types achieve significant improvements prediction tasks over other methods. can be applied archived unlocking valuable information facilitating new biomarker discoveries.

Language: Английский

Foundation models in bioinformatics DOI Creative Commons
Fei Guo, Renchu Guan, Yaohang Li

et al.

National Science Review, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 25, 2025

With the adoption of foundation models (FMs), artificial intelligence (AI) has become increasingly significant in bioinformatics and successfully addressed many historical challenges, such as pre-training frameworks, model evaluation interpretability. FMs demonstrate notable proficiency managing large-scale, unlabeled datasets, because experimental procedures are costly labor intensive. In various downstream tasks, have consistently achieved noteworthy results, demonstrating high levels accuracy representing biological entities. A new era computational biology been ushered by application FMs, focusing on both general specific issues. this review, we introduce recent advancements employed a variety including genomics, transcriptomics, proteomics, drug discovery single-cell analysis. Our aim is to assist scientists selecting appropriate bioinformatics, according four types: language vision graph multimodal FMs. addition understanding molecular landscapes, AI technology can establish theoretical practical for continued innovation biology.

Language: Английский

Citations

1
Towards the Next Generation of Data‐Driven Therapeutics Using Spatially Resolved Single‐Cell Technologies and Generative AI DOI Creative Commons

Avital Rodov,

Hosna Baniadam,

Robert Zeiser

et al.

European Journal of Immunology, Journal Year: 2025, Volume and Issue: 55(2)

Published: Feb. 1, 2025

ABSTRACT Recent advances in multi‐omics and spatially resolved single‐cell technologies have revolutionised our ability to profile millions of cellular states, offering unprecedented opportunities understand the complex molecular landscapes human tissues both health disease. These developments hold immense potential for precision medicine, particularly rational design novel therapeutics treating inflammatory autoimmune diseases. However, vast, high‐dimensional data generated by these present significant analytical challenges, such as distinguishing technical variation from biological or defining relevant questions that leverage added spatial dimension improve understanding tissue organisation. Generative artificial intelligence (AI), specifically variational autoencoder‐ transformer‐based latent variable models, provides a powerful flexible approach addressing challenges. models make inferences about cell's intrinsic state effectively identifying patterns, reducing dimensionality modelling variability datasets. This review explores current landscape technologies, application generative AI analysis their transformative impact on By combining with advanced methodologies, we highlight insights into pathogenesis disorders outline future directions leveraging achieve goal AI‐powered personalised medicine.

Language: Английский

Citations

0
Unified integration of spatial transcriptomics across platforms DOI Creative Commons

Eldad Haber,

Ajinkya Deshpande, Jian Ma

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: April 5, 2025

Spatial transcriptomics (ST) has transformed our understanding of tissue architecture and cellular interactions, but integrating ST data across platforms remains challenging due to differences in gene panels, sparsity, technical variability. Here, we introduce LLOKI, a novel framework for imaging-based from diverse without requiring shared panels. LLOKI addresses integration through two key alignment tasks: feature technologies batch datasets. Feature constructs graph based on spatial proximity expression propagate features impute missing values. Optimal transport adjusts sparsity match scRNA-seq references, enabling single-cell foundation models such as scGPT generate unified features. Batch then refines scGPT-transformed embeddings, mitigating effects while preserving biological Evaluations mouse brain samples five different demonstrate that outperforms existing methods is effective cross-technology program identification slice alignment. Applying ovarian cancer datasets, identify an integrated indicative tumor-infiltrating T cells Together, provides robust cross-platform studies, with the potential scale large atlas deeper insights into organization environments.

Language: Английский

Citations

0
New horizons at the interface of artificial intelligence and translational cancer research DOI
Josephine Yates, Eliezer M. Van Allen

Cancer Cell, Journal Year: 2025, Volume and Issue: 43(4), P. 708 - 727

Published: April 1, 2025

Language: Английский

Citations

0
Foundation models for bioinformatics DOI Open Access
Ziyu Chen,

Lin Wei,

Ge Gao

et al.

Quantitative Biology, Journal Year: 2024, Volume and Issue: 12(4), P. 339 - 344

Published: July 24, 2024

Abstract Transformer‐based foundation models such as ChatGPTs have revolutionized our daily life and affected many fields including bioinformatics. In this perspective, we first discuss about the direct application of textual on bioinformatics tasks, focusing how to make most out canonical large language mitigate their inherent flaws. Meanwhile, go through transformer‐based, bioinformatics‐tailored for both sequence non‐sequence data. particular, envision further development directions well challenges models.

Language: Английский

Citations

3
scProAtlas: an atlas of multiplexed single-cell spatial proteomics imaging in human tissues DOI Creative Commons
Tiangang Wang,

Xuanmin Chen,

Yujuan Han

et al.

Nucleic Acids Research, Journal Year: 2024, Volume and Issue: 53(D1), P. D582 - D594

Published: Nov. 11, 2024

Spatial proteomics can visualize and quantify protein expression profiles within tissues at single-cell resolution. Although spatial only detect a limited number of proteins compared to transcriptomics, it provides comprehensive information with By studying the distribution cells, we clearly obtain context multiple scales. includes composition cell types, functional structures, communication between regions, all which are crucial for patterns cellular distribution. Here, constructed annotation knowledgebase, scProAtlas (https://relab.xidian.edu.cn/scProAtlas/#/), is designed help users comprehensively understand different tissue types resolution across contains modules, including neighborhood analysis, proximity analysis network, construct maps multi-modal integration, gene identification, cell-cell interaction pathway display variable genes. data from eight imaging techniques 15 detailed 17 468 394 cells 945 region interests. The aim offer new insight into structure various annotation.

Language: Английский

Citations

2
Review: Single Cell Advances in investigating and understanding Chronic Kidney Disease and Diabetic Kidney Disease DOI
Sagar Bhayana, Philip Andreas Schytz, Emma T. B. Olesen

et al.

American Journal Of Pathology, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 1, 2024

Language: Английский

Citations

1
Identifying Differential Spatial Expression Patterns across Different Slices, Conditions and Developmental Stages with Interpretable Deep Learning DOI Creative Commons
Yan Cui, Zhiyuan Yuan

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 7, 2024

Abstract Spatially resolved transcriptomics technologies enable the mapping of multiplexed gene expression profiles within tissue contexts. To explore spatial patterns in complex tissues, computational methods have been developed to identify spatially variable genes single slices. However, there is a lack designed with differential (DSEPs) across multiple slices or conditions, which becomes increasingly common experimental designs. The challenges include complexity cross-slice and information modeling, scalability issues constructing large-scale cell graphs, mixed factors inter-slice heterogeneity. We propose DSEP identification as new task develop River, an interpretable deep learning-based method, solve this task. River comprises two-branch prediction model architecture post-hoc attribution method prioritize that explain condition differences. River’s special design for modeling spatial-informed makes it scalable omics datasets. proposed strategies decouple non-spatial components outcomes. validated performance using simulated datasets applied genes/proteins diverse biological contexts, including embryo development, diabetes-induced alterations spermatogenesis, lupus-induced splenic changes. In human triple-negative breast cancer dataset, identified generalizable survival-related DSEPs, unseen patient groups. does not rely on specific data distribution assumptions compatible various types, making versatile analyzing architectures conditions.

Language: Английский

Citations

0
PathOmCLIP: Connecting tumor histology with spatial gene expression via locally enhanced contrastive learning of Pathology and Single-cell foundation model DOI Creative Commons
Yong‐Ju Lee, Xinhao Liu, Minsheng Hao

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 11, 2024

Abstract Tumor morphological features from histology images are a cornerstone of clinical pathology, diagnostic biomarkers, and basic cancer biology research. Spatial transcriptomics, which provides spatially resolved gene expression profiles overlaid on images, offers unique opportunity to integrate features, thereby deepening our understanding tumor biology. However, spatial transcriptomics experiments with patient samples in either trials or care costly challenging, whereas generated routinely available for many legacy prospective cohorts disease progression outcomes well-annotated cohorts. Inferring computationally these would significantly expand biology, but paired data training multi-modal spatial-histology models remains limited. Here, we tackle this challenge by incorporating performant foundation pre-trained massive datasets pathology single-cell RNA-Seq, respectively, provide useful embeddings underpin models. To end, developed PathOmCLIP, model trained contrastive loss create joint-embedding space between histopathology RNA-seq model. We incorporate set transformer gather localized neighborhood architecture following training, further enhances performance is necessary obtain robust results. validate PathOmCLIP across five types achieve significant improvements prediction tasks over other methods. can be applied archived unlocking valuable information facilitating new biomarker discoveries.

Language: Английский

Citations

0