Breast cancer (BC) in patients with germline mutations ofBRCA1/BRCA2are associated with benefit from drugs targeting DNA damage response (DDR), but they account for only 5–7% of overall breast cancer. To define the characteristics of these tumors and also to identify tumors without BRCA mutation but with homologous recombination deficiency (HRD) is clinically relevant. To define characteristic features of HRD tumors and analyze the correlations betweenBRCA1/BRCA2and BC subtypes, we analyzed 981 breast tumors from the TCGA database using the signature analyzer. The BRCA signature was strongly associated with the HRD score top 10% (score ≥ 57) population. This population showed a high level of mutations in DDR genes, includingBRCA1/BRCA2. HRD tumors were associated with high expression levels ofBARD1andBRIP1. Besides,BRCA1/2 mutations were dominantly observed in basal and luminal subtypes, respectively. A comparison of HRD features in BC revealed thatBRCA1exerts a stronger influence inducing HRD features thanBRCA2does. It reveals genetic differences betweenBRCA1andBRCA2and provides a basis for the identification of HRD and other BRCA-associated tumors.
Key Points: Plasma proteome profiling identified distinct signatures across biopsy-proven primary glomerular disease subtypes. An elastic net model using 93 proteins classified primary glomerular disease subtypes and controls, with external validation. Integrating proteomics with machine learning yields biologically interpretable insights in primary glomerular diseases. Background: Primary GN is a heterogeneous group of kidney disorders where understanding of their pathophysiology remains incomplete. Despite the diagnostic potential of high-throughput proteomics, constrained proteomic depth and a reliance on binary comparisons have left the feasibility of using systemic signatures to differentiate multiple GN subtypes largely unexplored. Methods: To identify protein signatures that noninvasively differentiate major primary glomerular disease subtypes and provide mechanistic insights, we performed large-scale systemic proteome profiling of 5416 plasma proteins via Olink Explore HT in a discovery cohort ( n =147) and an external validation cohort ( n =85) of Korean participants (mean age, 41±13 years; 46% female). The study population included patients with four GN subtypes—focal segmental glomerulosclerosis, IgA nephropathy, minimal change disease, and membranous nephropathy—alongside healthy controls. We developed a machine learning (ML) model using logistic regression with elastic net regularization to classify disease groups based on proteomic profiles and evaluated its performance in the independent validation cohort. Results: Plasma proteome profiles were distinct among disease subtypes, emerging as a significant source of data variation independent of conventional markers such as eGFR or proteinuria levels. The ML model performed robustly in both the discovery and validation cohorts, achieving an area under the receiver operating characteristic curve >0.8 for differentiating minimal change disease, membranous nephropathy, and IgA nephropathy. The model, even without clinical information, correctly identified 93% of minimal change disease cases (14 of 15) and 63% of IgA nephropathy cases (20 of 32), but its performance was limited for focal segmental glomerulosclerosis, with only 21% of cases (three of 14) correctly classified. Functional analysis of key proteins highlighted distinct biologic pathways, such as hemostasis in minimal change disease. Conclusions: We identified distinct systemic proteome signatures for primary glomerular diseases, where disease subtype served as a major determinant of proteomic variance alongside conventional clinical markers. ML models demonstrated robust discriminatory performance for minimal change disease, membranous nephropathy, and IgA nephropathy, underscoring the potential for proteome-based classification.
Linking cellular states to clinical phenotypes is a major challenge in single-cell analysis. Here, we present single-cell multiple instance learning for sample classification and associated subpopulation discovery (scMILD), a weakly supervised multiple instance learning framework that robustly identifies condition-associated cells using only sample-level labels. After systematically validating scMILD’s accuracy through controlled simulations, we applied it to diverse disease datasets, confirming its ability to retrieve known biological signatures. Building on this, our sample-informed analysis of scMILD-identified monocytes in COVID-19 revealed a temporal transition from an early antiviral to a late stress-response state. Furthermore, in a cross-disease application, a model trained on COVID-19 successfully stratified patients with Lupus and distinguished shared inflammatory states from disease-specific ones. scMILD thus provides a validated and versatile strategy to dissect cellular heterogeneity, bridging single-cell observations with high-level phenotypes.
Foundation models (FMs) pre-trained on large-scale single-cell RNA sequencing (scRNA-seq) data provide powerful cell embeddings, but their practical usability and systematic comparison are limited by model-specific environments, preprocessing pipelines, and execution procedures. To address these challenges, we introduce scUnify, a unified zero-shot inference framework for single-cell foundation models. scUnify accepts a standard AnnData object and automatically manages environment isolation, preprocessing, and tokenization through a registry-based modular design. It employs a hierarchical distributed inference strategy that combines Ray-based task scheduling with multi-GPU data-parallel execution via HuggingFace Accelerate, enabling scalable inference on datasets containing up to one million cells. In addition, built-in integration of scIB and scGraph metrics enables standardized cross-model embedding evaluation within a single workflow. Benchmarking results demonstrate substantial reductions in inference time compared with the original model implementations, while preserving embedding quality and achieving near-linear multi-GPU scaling. scUnify is implemented in Python and is publicly available at https://github.com/DHKim327/scUnify .
SUMMARY: Linking cellular states to clinical phenotypes is a major challenge in single-cell analysis. Here, we present scMILD, a weakly supervised Multiple Instance Learning framework that robustly identifies condition-associated cells using only sample-level labels. After systematically validating scMILD’s accuracy through controlled simulations, we applied it to diverse disease datasets, confirming its ability to retrieve known biological signatures. Building on this, our sample-informed analysis of scMILD-identified monocytes in COVID-19 revealed a temporal transition from an early antiviral to a late stress-response state. Furthermore, in a novel cross-disease application, a model trained on COVID-19 successfully stratified Lupus patients and distinguished shared inflammatory states from disease-specific ones. scMILD thus provides a validated and versatile strategy to dissect cellular heterogeneity, bridging single-cell observations with high-level phenotypes.