biostudies-arrayexpressDavid JohnHomo sapienshttps://www.ebi.ac.uk/biostudies/studies/E-MTAB-14753Single-cell transcriptomics has emerged as a powerful technology for understanding cardiovascular diseases, providing valuable insights into transcriptomic changes linked to heart failure, both with and without preserved ejection fraction. However, significant gaps persist in our understanding of the molecular mechanisms underlying these different types of heart failure, and we continue to seek innovative approaches to tackle the challenges posed by the increasing scale and complexity of single-cell datasets in identifying meaningful gene signatures. The integration of machine learning, particularly deep neural networks, has shown immense promise in addressing these challenges by learning transcriptional patterns from single-cell transcriptomic data, reconstructing expression profiles, and effectively classifying cells. Recent advancements in explainable artificial intelligence enhanced the interpretation of these models by attributing importance scores, such as Shapley values. However, methods for identifying differentially regulated gene contribution scores have not yet been integrated. In this study, we introduce a novel method to identify differentially explained genes (DXGs) based on importance scores derived from custom-built neural networks capable of classifying heart failure subtypes at the single-cell level. We highlight the superiority of DXGs in identifying heart failure-relevant pathways, presenting them in a format comparable to traditional differential expression analyses. Using this method, we identify novel signatures providing new insights into the molecular basis of heart failure and offering a robust foundation for future research and therapeutic exploration.biostudies-arrayexpressSample Collection - For Mouse hearts, the following protocol was used: cardiac tissues were thawed on ice and cut into small pieces. Minced tissue was pre-digested with a 5-ml enzyme solution of collagenase (2,500 U, Thermo Fisher Scientific) in HBSS+/+ (Gibco) for 10 min at 37 °C in a water bath. After centrifugation at 500g and 4 °C for 5 min, the supernatant was discarded, and nuclei were isolated after cell disruption with a glass dounce homogenizer (five strokes with a loose pestle and ten strokes with a tight pestle)40,41. After filtering (20-µm strainer, pluriSelect), the suspension was centrifuged at 1,000g and 4 °C for 6 min and resuspended in 500 µl staining buffer containing 1% BSA (Sigma-Aldrich), 5 nM MgCl2 (Sigma-Aldrich), 1 mM EDTA (Gibco), 1 mM EGTA (Gibco), 0.2 U µl−1 RNasin Plus Inhibitor (Promega) and 0.1 µg ml−1 Hoechst (Life Technologies) in Dulbecco´s phosphaste buffered saline (DPBS). Hoechst-positive nuclei were separated from cell debris by using the FACSAria Fusion instrument (BD Biosciences) and sorted into staining buffer without Hoechst at 4 °C. For the two human heart failure samples, biopsies were taken during surgery of the aortic valve, with tissue from the hypertrophied interventricular septum in the left ventricle being collected (75-250 mg per sample; Department of Cardiovascular Surgery, University Hospital Frankfurt am Main, Germany). Informed consent was obtained from all two patients. The study was approved by an institutional review committee of the University Hospital of the Johann Wolfgang Goethe University in compliance with internal standards of the German government and procedures followed were in accordance with institutional guidelines (Application 347/18) and the Declaration of Helsinki.Nucleic Acid Extraction - Prior to snRNA-seq, single nuclei were isolated from the hypertrophied interventricular septum in the left ventricle. Cardiac tissues were thawed on ice and cut into small pieces. Minced tissue was then pre-digested in 5 ml enzyme solution of collagenase (2500 U, Thermo Fisher Scientific) in HBSS+/+ (Gibco) for 10 min at 37 °C in a water bath. After centrifugation at 500x g and 4 °C for 5 min, the supernatant was discarded and nuclei were isolated after cell disruption with a glass dounce homogenizer, according to a previously published protocol40. After homogenization of the tissue and filtering through a 20 µm strainer (pluriSelect), the nuclei suspension was centrifuged at 1,000x g at 4°C for 6 min and resuspended in 500 µl staining buffer containing 1 % BSA (Sigma Aldrich), 5 nM MgCl2 (Sigma Aldrich), 1 mM EDTA (Gibco), 1 mM EGTA (Gibco), 0.2 U/µl RNasin® Plus Inhibitor (Promega) and 0.1 µg/ml Hoechst (Life Technologies) in DDPBS. Then, Hoechst-positive single nuclei were separated from cell debris by FACS using the FACS ARIA Fusion instrument (BD Biosciences) and sorted into staining buffer without Hoechst at 4 °C.Sequencing - Library quantification and quality assessment were performed using Bioanalyzer Agilent 2100 using a High Sensitivity DNA chip (Agilent Genomics). Indexed libraries were equimolarly pooled and sequenced on Illumina NovaSeq 6000 using paired-end 26x98 bp as sequencing mode by GenomeScan (Leiden, Netherlands).Library Construction - Nuclei suspensions were loaded on a 10X Chromium Controller (10X Genomics) according to manufacturer’s protocol based on the 10X Genomics proprietary technology. All snRNA-seq libraries were prepared using Chromium Single Cell 3′ v3 Reagent Kit (10X Genomics) according to manufacturer’s protocol. Briefly, the initial step consisted in performing an emulsion where individual nuclei were isolated into droplets together with gel beads coated with unique primers bearing 10X cell bar-codes, UMI (unique molecular identifiers) and poly(dT) sequences. Reverse transcription reactions were engaged to generate barcoded full-length cDNA followed by the disruption of emulsions using the recovery agent and cDNA clean up with DynaBeads MyOne Silane Beads (Thermo Fisher Scientific). Total cDNA was amplified using a Biometra Thermocycler TProfessional Basic Gradient with 96-Well sample block (98 °C for 3 min; cycled 14×: 98 °C for 15 s, 67 °C for 20 s, and 72 °C for 1 min; 72 °C for 1 min; held at 4 °C). Amplified cDNA product was cleaned with the SPRIselect Reagent Kit (Beckman Coulter). Indexed sequencing libraries were constructed using the reagents from the Chromium Single Cell 3′ v3 Reagent Kit, as follows: fragmentation, end repair and A-tailing; size selection with SPRIselect; adaptor ligation; post-ligation cleanup with SPRIselect; sample index PCR and cleanup with SPRI select beads.OrganizationMINSEQE ScoreAssays and DataProcessed DataMAGE-TAB FilesData Transformation - The secondary data analysis was initiated by using the Seurat 4.1.0 package in R. The data sets were first combined into a Seurat object and then subjected to a filtering process. Barcodes with too low (< 300) or too high number of genes (> 6000) were sorted out and not considered further in the data analysis. In addition, barcodes with too low (< 500) and too high read counts (> 15000) were also sorted out. To further ensure that no apoptotic cells or doublets were analyzed, we discarded barcodes with a high percentage of mitochondrial content (> 5%). The filtered gene counts were then logarithmized and normalized according to the tutorial for data analysis with Seurat.Sequence Alignment - Single-cell RNA-seq results were processed by CellRanger (10x Genomics) version 7.0.0 software. The first step consisted of demultiplexing and processing raw base count files by the implemented mkfastq tool. The human raw reads were mapped to the reference genome hg38 (GRCh38-2020) using Cellranger count, whereas the mouse raw reads were mapped to the reference genome mm10 (GRCm38-2020).MetabolomicsUnknownTranscriptomicsGenomicsProteomicsIllumina NovaSeq 6000RNA-seq of coding RNA from single cellsHomo sapiensDavid JohnfalseSingle nuclei sequencing of human HFpEF (Heart Failure with reduced ejection fraction) and Mice TAC (Transaortic constriction) dataSingle-cell transcriptomics has emerged as a powerful technology for understanding cardiovascular diseases, providing valuable insights into transcriptomic changes linked to heart failure, both with and without preserved ejection fraction. However, significant gaps persist in our understanding of the molecular mechanisms underlying these different types of heart failure, and we continue to seek innovative approaches to tackle the challenges posed by the increasing scale and complexity of single-cell datasets in identifying meaningful gene signatures. The integration of machine learning, particularly deep neural networks, has shown immense promise in addressing these challenges by learning transcriptional patterns from single-cell transcriptomic data, reconstructing expression profiles, and effectively classifying cells. Recent advancements in explainable artificial intelligence enhanced the interpretation of these models by attributing importance scores, such as Shapley values. However, methods for identifying differentially regulated gene contribution scores have not yet been integrated. In this study, we introduce a novel method to identify differentially explained genes (DXGs) based on importance scores derived from custom-built neural networks capable of classifying heart failure subtypes at the single-cell level. We highlight the superiority of DXGs in identifying heart failure-relevant pathways, presenting them in a format comparable to traditional differential expression analyses. Using this method, we identify novel signatures providing new insights into the molecular basis of heart failure and offering a robust foundation for future research and therapeutic exploration.2025-10-01T00:00:00Z2025-10-01T01:06:00.39Z2025-01-17T18:51:16.234ZE-MTAB-14753ERP168423E-MTAB-11268E-MTAB-13264E-MTAB-7869EFO_0002944EFO_0004170EFO_0005684EFO_0004917EFO_0005518EFO_0003816EFO_0004184