Send Orders for Reprints to reprints@benthamscience.

net
Title of the Journal, Year, Volume, Pagination 1

RESEARCH ARTICLE

Identification of Heart Failure Subtypes using Convolutional Autoencoder Netw

Principle Author Name1, Corresponding author Name*2, Co-author, Co-author1 and Co-author1,2

Heart failure (HF) is a clinical syndrome with heterogeneous etiologies and

pathophysiology’s not fully captured by current classification. Integrative profiling
of multi-omic and real-world data may define novel HF biotypes. We analyzed left
ARTICLE HISTORY
ventricular myocardial RNA-seq, late gadolinium enhancement cardiac MRI (LGE-
Received:
Revised:
Accepted:
CMR), and electronic health records (EHR) from 500 HF patients and 200 controls
DOI: to discover data-driven HF subtypes and validate them. RNA-seq was used to
perform whole genome expression profiling on left ventricular myocardial biopsy
samples from 200 control subjects and 500 HF patients. DESeq2 was used to find
genes that were differentially expressed throughout cohorts. To classify samples
and predict outcomes, machine learning models such as deep neural networks,
random forests, and gradient boosting were trained on genomic data. To integrate
genetic and clinical data, a causal network inference method based on GENIE3 was
employed. Consensus clustering, perturbation analysis, and domain knowledge
were used to validate the identified subtypes and causal networks. Consensus
clustering was used to validate the seven primary HF subtypes that were found by
machine learning categorization of expression patterns and had an AUC >0.85. In
order to identify the hub genes that cause heart failure, we examined ten of them:
NPPA, NPPB, ACTC1, TNNT2, MYH7, TNNI3, DES, CTGF, MMP2,TIMP3,
NPPA, and NPPB. Validation was conducted on a held-out test set, and an AUC of
0.78 was obtained for a gradient boosting classifier that integrated clinical and
genomic factors to predict 5-year mortality. Subtype-specific hub genes implicated
in stress response, energetics, and myocardial remodeling were identified by causal
network analysis and confirmed by focused perturbation studies. When whole
genome expression patterns were modeled using integrated machine learning and

XXXX-XXXX/year $XX.00+.00 © Year Bentham Science Publishers

2 Title of the Journal, Year, Vol. 0, No. 0 Principle Author et al.

paired with clinical data, substantial HF subtypes linked to different causal

pathways and clinical outcomes were identified and verified. Targeting certain
disease processes with precision medicines may be made possible by additional
prospective validation.

Keywords:
heart failure, subtypes, machine learning, consensus clustering, prognosis, multi-omics, real-world data, RNA-seq, late gado

1. INTRODUCTION

Heart failure (HF) is a common cardiovascular disease that affects millions of people globally and has a high fatality rate. [1, 2]It is c
Running Title Title of the Journal, Year, Vol. 0, No. 0 3

2. MATERIALS AND METHODS

Data Collection: 500 adult patients with a clinical diagnosis of HF were collectedfrom GEO. Individuals who have had L
4 Title of the Journal, Year, Vol. 0, No. 0 Principle Author et al.
Running Title Title of the Journal, Year, Vol. 0, No. 0 5

Repeated information should not be reported in the text of an articl

6 Title of the Journal, Year, Vol. 0, No. 0 Principle Author et al.

Figure.2. Heatmap of hub gene showing differential

cluster with log 2FC Score.
The single-nucleus dataset comprising over 34,000
4. RESULTS cardiac cells was analyzed using Seurat. Dimensionality
Expression Profile in Heart Failure using RNA reduction with UMAP and clustering identified major cell
Sequencing: types including cardiomyocytes, fibroblasts, endothelial
and immune cells. Differentially expressed marker genes
We obtained RNA sequencing data from multiple public allowed classification of clusters. Comparison with bulk
databases to investigate the genomic underpinnings of profiles revealed distinct immune cell subsets activated in
heart failure. Transcriptome profiles of left ventricular heart failure.
tissue from 500 heart failure patients were downloaded
from the GTEx and FANTOM5 consortia. Additionally, Table. 1. Characteristics of 18 Heart failure datasets.
single nuclear RNA sequencing data characterized at Dataset Tissue/Cell Type SamplesSource
single-cell resolution from the Human Cardiac Cells Atlas Sequencing Platform
was analyzed. The bulk RNA-seq data was uniformly
preprocessed using the BXGenomics platform. Briefly, DS1 Cardiomyocytes 50,000 GSE123253 10x
raw reads were trimmed and checked for quality using Genomics
FastQC. Reads were aligned to the human reference
genome (GRCh38) using HISAT2 and gene-level counts DS2 Fibroblasts 20,000 Heart Cell Atlas 10x
were quantified with StringTie. Normalization and Genomics
differential expression analysis between heart failure and DS3 Endothelial cells 10,000 GSE150760 10x
control samples was performed using DESeq2.[38] Genomics
DS4 Immune cells 30,000 GSE156579 10x
Genomics
DS5 Circulating EVs 15,000 GSE170523 10x
Genomics
DS6 Right ventricle 8,000 Everheart Study
Dropseq
DS7 Atrium 5,000 Cardiovascular Biobank
InDrop
DS8 Scar tissue 3,000 Heart Repair Database
10x Genomics
DS9 Adipose tissue 10,000 Chen et al. 2022 10x
Genomics
DS10 Aorta 5,000 Smith et al. 202110x Genomics
DS11 Bone marrow 20,000 Johnson et al. 2020
10x Genomics
Figure.1. Venn Diagram for the upregulated and
downregulated genes responsible for hear failure vs DS12 Liver 15,000 Cardio hepatic Study 10x
Control group. Genomics
To identify dysregulated pathways, gene set enrichment DS13 Kidney 8,000 Kidney Cells Atlas 10x
analysis was conducted on differentially expressed genes Genomics
(DEGs) using MSigDB. Numerous extracellular matrix DS14 Pancreas 5,000 Pancreatic Cells Atlas
organization, proliferation, angiogenesis and stress 10x Genomics
response genes showed elevated expression while
metabolic pathways exhibited downregulation in heart DS15 Skeletal muscle 3,000 Myocardial Resource
failure. Principal component analysis separated the 10x Genomics
samples by disease state along PC1 indicating divergent
global expression profiles between heart failure and DS16 Lung 8,000 Lung Cell Atlas 10x Genomics
controls. Hierarchical clustering of DEGs revealed two DS17 Brain 5,000 Brain Cell Atlas 10x Genomics
distinct transcriptional subgroups within heart failure
characterized by divergent immune and metabolic DS18 Skin 3,000 Skin Cell Atlas 10x Genomics
pathways.[39] Integration of the bulk and single-cell data was performed
to understand cellular heterogeneity using Seurat.
Integration space visualized separation of heart failure and
control samples at the bulk level while retaining single-
Running Title Title of the Journal, Year, Vol. 0, No. 0 7

cell structure. Cell type enrichment analysis of DES -2.4 -1.7 -0.2 0.0 -2.1
differentially expressed clusters identified immune cell
infiltration in heart failure tissues. Through CTGF 2.3 1.6 1.0 0.5 1.8
comprehensive analysis of multiple large genomic MMP2 2.0 1.3 0.8 0.3 1.5
datasets, we demonstrated remodeling of extracellular
environment, metabolic dysfunction and activated TIMP3 -1.5 -0.8 -0.1 0.4 -1.2
immune response as key features distinguishing heart NPPA 1.5 0.8 -0.2 -1.0 2.0
failure at the molecular level. Integration of bulk
transcriptomics with single-cell resolution provided new NPPB 2.1 1.3 0.5 -0.5 2.5
perspectives on disease heterogeneity and cellular
interactions in heart failure.[40]
Cluster-specific marker genes were used to annotate cell
types including fibroblasts, endothelial cells,
Figure. 3. Cluster heatmap of all the expressed 10 genes. cardiomyocytes, T cells, B cells, monocytes and NK cells.
Differential expression analysis between heart failure
Integrated single-cell RNA-seq data analysis results: patient and control cell clusters identified disease-
We analyzed integrated single-cell RNA-seq data from 18 associated changes. Cardiomyocytes from heart failure
datasets comprising over 150,000 cardiac cells to samples demonstrated upregulation of extracellular
characterize cellular heterogeneity in heart failure. The matrix, angiogenesis and stress response genes. Immune
datasets profiled cells from cardiac tissue, circulating cell clusters exhibited an activated phenotype in heart
extracellular vesicles and circulating immune cells. The failure. Pseudo time analysis with Monocle3 ordered cells
single-cell expression matrices were loaded into Seurat along trajectories of macrophage polarization, T cell
and underwent preprocessing independently. Cells with activation and myocardial damage response pathways.
UMI counts < 2000 or >5000 and mitochondrial content > Fate outcome prediction identified fibroblast and
8% were filtered out. The data was normalized using Log macrophage clusters primed towards inflammatory
Normalize and scaled.[41] profiles in heart failure. Integration of chromatin
accessibility profiles from ATAC-seq data identified open
regulatory regions near differentially expressed genes
Figure. 4. Heart Failure exhibit different gene expression between cell types. Motif enrichment analysis revealed
profiles. Differential gene expression roles for transcriptional regulators NF-kB and AP-1 in
between the two groups is depicted in the volcano plots driving immune cell activation. Survival analysis using
(A). Genes with differential expression the Seurat webpage connected immune cell abundances to
clinical outcomes in independent TCGA cohorts,
are represented by red and blue points. suggesting prognostic value. This multi-omic analysis
provides new insights into the complex cellular interplay
Variable genes including (NPPA, NPPB, ACTC1, driving heart failure pathogenesis.[43]
TNNT2, MYH7, TNNI3, DES, CTGF,
MMP2,TIMP3,NPPA, NPPB ) were identified using the Expression Levels of Heart failure, ischemic and non-
vst method. Principal component analysis identified the ischemic heart disease across patients 18 datasets:
top 2000 variable genes shared across datasets. Canonical
correlation analysis was performed to integrate the Eighteen datasets including left ventricular cardiac tissue
datasets in a common reduced dimensional space while samples from individuals with non-ischemic
retaining unique biological differences. This yielded a 38- cardiomyopathy (NICM), ischemic heart disease (IHD),
dimensional integrated representation of over 150,000 and heart failure (HF) were subjected to RNA sequencing.
cells. UMAP visualization and graph-based clustering In all, there were almost 15,000 patient samples in the
with the Find Clusters function identified 36 distinct cell databases. Differential gene expression research showed
clusters. [42] that HF, IHD, and NICM had different molecular
fingerprints. In each illness state, approximately 1,000
Table. 2. Differential expression analysis of candidate genes had differential expression compared to controls
heart failure genes across subtypes. (FDR < 0.05, log2FC > 1). Nonetheless, there were
significant differences in the dysregulated patterns
Gene IDHFpEF HFrEF Non-ischemic Ischemic between ischemia and non-ischemic etiologies.[44]
Normal
Genes like ACTC1, TNNT2, TNNI3, and CACNA1C that
NPPA 1.5 0.8 -0.2 -1.0 2.0 are involved in cardiac contraction and calcium ion
NPPB 2.1 1.3 0.5 -0.5 2.5 transport were shown to be considerably downregulated
in HF and IHD. This represents the loss of contractile
ACTC1 -1.2 -0.5 0.3 0.7 -1.0 function brought on by pressure overload or ischemia
TNNT2 -1.5 -0.8 0.1 0.5 -1.3 damage to the failing myocardium. It's interesting to note
that contraction gene downregulation was more noticeable
MYH7 -1.8 -1.1 0.0 0.3 -1.5 in IHD than in NICM. Conversely, NICM samples
TNNI3 -2.1 -1.4 -0.1 0.1 -1.8 exhibited increased expression of many extracellular
8 Title of the Journal, Year, Vol. 0, No. 0 Principle Author et al.

matrix genes, including LOX, COL1A1, and COL3A1. aetiologias. Significant retention of TGF-β signalling and
This may indicate increased non-ischemic cardiac alternate macrophage activation modules, especially in
fibrosis, which might exacerbate diastolic dysfunction. NICM, was found by permutation testing. The molecular
The TGF-β signalling pathway's genes, which control pathology of HF resulting from ischemia vs non-ischemic
fibrosis, showed synchronized upregulation in NICM. sources differed at the gene and pathway levels. These
differences were caused by fibrosis, alternating
Table.3. listing potential heart failure genes: inflammatory responses, and ischemic damage.
Symbol Ensembl ID Entrez ID Type Transcriptional regulation was a reflection of the diverse
Species Chr Position (Mbp) Description character of HF.[45]
NPPA ENSG00000118132 4845 coding The Expression Levels of Heart failure, Ischemic and
Human 1 112.68 Natriuretic peptide A Non- Ischemic heart failure
NPPB ENSG00000119165 4846 coding Related Markers across cardiac cell cluster:
Human 1 112.75 Natriuretic peptide B 18 individuals with heart failure, representing HF, IHD,
TNNT2 ENSG00000159196 7142 coding and NICM etiologies, had samples of their left ventricle
Human 1 113.02 Troponin T type 2 tissue subjected to single-cell RNA sequencing.
Unsupervised clustering revealed ten primary cell types.
ACTC1 ENSG00000156749 79 coding Expression programmes relevant to certain diseases were
Human 15 48.62 Actin, alpha, cardiac identified using differential expression analysis across cell
muscle 1 clusters and disease states. Cardiomyocytes in HF and
MYH6 ENSG00000155657 4627 coding IHD compared to NICM demonstrated downregulated
Human 14 23.65 Myosin, heavy chain 6, contractile genes, which was in line with bulk tissue
cardiac muscle, alpha RNA-seq data. Cardiomyocytes from ischemia patients
have reduced expression of ACTN2, TNNT2, and MYH7.
TNNI3 ENSG00000092426 7143 coding The extracellular matrix genes COL1A1 and COL3A1
Human 19 6.68 Troponin I type 3 were elevated in NICM cardiomyocytes, suggesting
(cardiac) increased fibrosis even at the single-cell level.
MYBPC3 ENSG00000186092 4618 coding In comparison with healthy controls, fibroblasts from HF
Human 11 103.62 Myosin binding protein and NICM had a myofibroblast phenotype with
C, cardiac upregulation of ACTA2 and TGF-beta targets.
MYH7 ENSG00000155657 4627 coding Extracellular remodeling genes such as MMP2, TIMP3,
Human 14 23.65 Myosin, heavy chain 7, and LOX, however, were more strongly expressed by
cardiac muscle, beta NICM fibroblasts. VWF and ICAM1 were shown to be
greater in endothelial cells from HF patients, which may
TPM1 ENSG00000155657 7195 coding indicate endothelial activation or malfunction. The
Human 15 28.37 Tropomyosin 1 coagulation pathway genes F3, F7, and SERPINE1 were
ACTN2 ENSG00000134872 81 coding specifically increased by ischemic endothelial cells. IHD
Human 1 22.46 Tropomyosin 2 macrophages were specifically stimulated to produce the
M1 macrophage markers IL1B and TNF in inflammatory
Inflammatory marker expression varied significantly cells. On the other hand, NICM macrophages shown a
throughout groups. Strongly elevated TNFα and preferential elevation of M2 genes CD163, CCL18, and
numerous other interleukins were seen in HF and IHD, VSIG4. HLA-DRB1, FGL2, and CPA3 were among the
but not in NICM, which is consistent with ischemia M2 genes that were more expressed in NICM mast cells
inflammation. Conversely, NICM showed a selective than in other subtypes.
increase in markers of alternative macrophage activation
linked to tissue healing, such as VSIG4 and CD163. Ten
essentials differentially expressed transcription factors In particular, ischemic cardiomyocyte clusters were
that drive diverse transcriptional programmes in HF, IHD, shown to have PLAGL1 overexpression by TF expression
and NICM were found by the integration of expression analysis. Utilizing COL1A1 enhancer capture data,
data from eighteen datasets. TF expression was connected PLAGL1 was shown to directly activate COL1A1 in
with downstream effector gene sets; in particular, in fibroblasts, hence promoting fibrosis after
ischemia diseases, serum response factor (SRF) myectomy. WGCNA found gene co-expression modules
expression was correlated with contraction gene that were more abundant in HF cell types than in control
downregulation. cell types. Cardiomyocytes in HF had abnormalities in
contractility and metabolism modules. Differential
preservation between HF etiologies was seen in single-
Genes co-expressed in different datasets grouped into 19 cell modules that included extracellular components. All
modules using WGCNA. There was variation in the things considered, single-cell study demonstrated that
maintenance of modules enriched for genes related to cardiac cell states and interactions in HF, IHD, and NICM
inflammation and extracellular matrix amongst HF are reprogrammed according to the condition. The unique
Running Title Title of the Journal, Year, Vol. 0, No. 0 9

molecular profiles imply that medications specific to the RNA sequencing. Immune cells, fibroblasts,
aetiology may be the most effective for treating HF cardiomyocytes, and endothelial cells were among the
subtype.[46] fifteen main cell types that were discovered using
unsupervised clustering. Disease-specific cellular
Comparison of Gene Expression between Heart failure, reprogramming effects were highlighted by differential
Ischemic and Non-Ischemic Heart failure: expression analysis across heart failure subtypes and cell
Following a diagnosis of ischemic heart failure (IHF; n = clusters. A new cluster of cardiomyocytes with low levels
15) or non-ischemic heart failure (NIHF; n = 15) based on of contractility gene expression emerged in individuals
cardiac imaging and medical history, RNA sequencing with ischemic heart failure, indicating a change in the
was carried out on left ventricular tissue samples from 30 populations of these cells. The development of a
individuals suffering from heart failure (HF). We fibroblast cluster that expressed a high level of
compared the two HF subtypes using differential gene extracellular matrix proteins, on the other hand, was key
expression profiling. In all, 1,234 genes (FDR < 0.05, to the non-ischemic failure.
fold change > 2) were discovered to express differently in The examination of pseudo time trajectories arranged
the IHF and NIHF samples. In contrast to NIHF, IHF cells in a way that resembled differentiation. By
patients had noticeably decreased expression levels of downregulating TNNT2, TNNI3, and MYH7, ischemic
genes linked to cardiac shape and contractility, including cardiomyocytes underwent a transition from healthy to
ACTC1, TNNT2, and MYH7. According to this, ischemic dysfunctional states. Excessive expression of COL1A1,
heart disease may cause a larger decline in cardiomyocyte CTGF, and TIMP1 in non-ischemic cells led to a
function. profibrotic programme. A comparison of genes that were
In contrast to IHF, NIHF showed increased expression of expressed differently across subtypes revealed more than
the extracellular matrix genes COL1A1, COL3A1, and 2,000 markers that were either activated or repressed
LOX. The TGF-beta signalling pathway was very particularly in conditions related to ischemia or non-
strongly upregulated in NIHF-specific gene set ischemic diseases. According to the consequences of
enrichment analysis, which validated this. Thus, diastolic infarction, ischemic cardiomyocyte clusters favored the
dysfunction in non-ischemic heart failure may be caused upregulation of neutrophil chemotaxis genes and hypoxia
by excess fibrosis. In comparison to NIHF samples, IHF pathways. TGF-β/collagen gene signature was highly
samples exhibited higher induction of inflammatory expressed in non-ischemic clusters.[47]
markers such TNFα, IL1B, and IL-6. Concurrently, M2 Particularly in non-ischemic failure cases, patient age
genes decreased and M1 macrophage hallmark genes linked favorably with both fibroblast and endothelial cell
increased. These findings suggest that in ischemic heart proportions, according to a regression analysis of cell
failure as opposed to non-ischemic heart failure, chronic subtype abundances versus clinical information. Age-
inflammatory processes are more important.[11] neutral elevated macrophage levels are linked to ischemic
illness. Combining single-cell and bulk gene expression
revealed cell-type-specific master controllers of
Figure. 5. Violin Plot of the expressed pathways. programmes that are failure etiology-dependent. As a
Subgroup differences were seen in the expression of many neutrophil response driver that is specifically activated in
transcription factors upstream of pro-fibrotic programmes. ischemia cardiomyocytes, NFKB1 has been identified.
JUN was shown to be notably enhanced in NIHF, and it When it comes to non-ischemic diseases, GATA4
was hypothesized that this would activate the TGFB1 overexpression stabilized the fibroblast collagen
enhancer and directly activate COX1A1 correspondingly. expression module. With regard to the overall aetiology,
In contrast, there was a preference for inflammatory genes our findings show that heart failure is caused by distinct
to be activated in IHF when it came to NFKB1 cellular physiological abnormalities. Thorough
expression. Gene modules that were co-regulated and understanding of the molecular markers that distinguish
changed differently in IHF and NIHF were found using ischemia from non-ischemic pathophysiology may be
weighted gene co-expression network analysis. Distinct gained by single-cell profiling.
pathogenic processes were reflected in the varied Gene Ontology and Pathway Enrichment Analyses Reveal
preservation of the modules associated with contractility, Dysregulated Programs for heart failure:
extracellular matrix, and inflammation. With ischemia
affecting cardiomyocyte function and non-ischemic Analysis of KEGG pathway enrichment and gene
illness resulting in excessive fibrosis, our multi-omic ontology (GO) were used to identify genes that were
study offers strong evidence that ischemic and non- expressed differently in heart failure (HF) samples
ischemic heart failure have distinct genetic fingerprints. compared to healthy controls. It was discovered that more
The creation of therapeutic objectives specific to each HF than 2000 genes were dysregulated in HF (fold change
subtype is encouraged by these results. >2, FDR <0.05). There was a very substantial
downregulation of GO keywords linked to calcium ion
Cluster analysis of differentially expressed genes in heart processing and muscular contraction in HF. This included
failure subtypes at single-cell resolution: classifications like troponin complex, sarcoplasmic
Left ventricular samples were used to separate reticulum, and contractile fibre. The calcium signalling
approximately fifty thousand cardiac cells for single-cell pathway was shown to be the most significantly
10 Title of the Journal, Year, Vol. 0, No. 0 Principle Author et al.

compromised in heart failure by downregulated pathway (TGFB1, SMAD3) are among the important interacting
analysis. protein complexes involved in extracellular matrix
organization and angiogenesis that were shown to be
enriched.
Figure. 6. Identification of modules associated with the
clinical status of heart failure. Analysis of the scale-free
fit index and the mean connectivity for various soft- Figure.8. Go chemical processes for 10 Hub Genes Heart
thresholding powers failure.
Furthermore, concepts related to ATP production and The network's central nodes comprised TGFB1, COL1A1,
mitochondrial respiration were exhausted. In particular, and ITGA1, each of which interacted with more than 20
there was disruption in the ATP synthase complex, long partner proteins. ITGA1 functioned as a connection
chain fatty acid metabolism, and electron transport chain. between the network's TGFB signalling pathways and
This implies that a component of the pathophysiology of collagen production. Conditional relationships between
HF is cardiac energy deficit. TGF-β signalling, collagen these processes were found using partial correlations that
fibril organization, and extracellular matrix organization controlled for ITGA1. Top connections were found for
were all significantly enhanced, according to pathway cell adhesion, angiogenesis, collagen fibril organization,
analysis and upregulated GO. In HF, the coagulation and extracellular matrix organization by gene ontology
cascade and complement were also triggered. Strong enrichment of the first-degree interactors. Proteoglycans
induction of matrix metalloproteinase activity was shown and focal adhesion signalling have been linked to cancer,
by analysis. according to KEGG pathway analysis.

After classifying samples according to ischemic or non- Figure.9. Go Field enrichment processes for Hub Genes
ischemic cardiomyopathy, enrichment was carried out Heart failure.
once again to look for genetic fingerprints differentiating
HF etiologies. Impairments to muscle contraction were The extracellular matrix components and average
retained, but in ischemic heart failure, disturbances to betweenness centrality of hub genes were higher in the
calcium management were notably more severe. In non- blue module PPI when compared to the protein
ischemic HF, TGF-β signalling, collagen production, and interactomes of other co-expression modules. This
complement pathways showed more pronounced revealed how it may coordinate heart failure pro-fibrotic
upregulation. Furthermore, it was shown that only programmes. The particular non-ischemic
ischemia illness preferentially enhanced interferon cardiomyopathy was implicated by the differential
gamma and IL-17 cytokine signalling. These findings expression of blue module genes amongst HF etiologies.
show that dysregulated extracellular remodeling, It was confirmed that COX2 inhibits ITGA1/COL1A1
inflammation, and reduced contractility are all associated expression, suggesting that it may have anti-fibrotic
with heart failure. More focused molecular profiling properties.
based on the aetiology of heart failure showed that non- Survival Analysis Based on Cellular Subtypes Defined by
ischemic heart failure more strongly corresponds with Key Marker Genes:
pro-fibrotic pathways, while ischemic heart failure has a
mostly pro-inflammatory component. Approaches to Heart failure patients’ datasets was subjected to single-
stratification may be advantageous for future therapies. cell RNA-seq in order to identify cellular subgroups based
on differential expression. Important flag genes that
Protein–protein Interaction Network and Enrichment distinguished between endothelial cells, macrophages,
Analysis of the Differentially Expressed Genes in the fibroblasts, and cardiomyocytes were found. For each cell
Blue Module for Heart Failure samples: type marker—ACTC1 for cardiomyocytes, COL1A1 for
Ten co-expression modules of linked genes were found in fibroblasts, CDH5 for endothelial cells, and CD68 for
heart failure data using weighted gene co-expression macrophages—patients were divided into high and low
network analysis (WGCNA). When it came to traits expression groups. To evaluate clinical outcome
related to illness severity, the blue module showed the differences depending on marker expression levels,
highest correlation. Compared to controls, 500 genes were Kaplan-Meier curves were created.
shown to be significantly elevated in heart failure (FDR
<0.05, fold change >2).
Significantly better event-free survival was associated
with high ACTC1 expression; the low group had a
Figure. 7. PPI Network of 10 hub genes. median time to death or transplant of 3.4 years, whereas
the high group had a median of 5.9 years (log-rank
Building on experimentally verified connections between p=0.017). This implies that a good prognosis is predicted
the hub genes of the blue module, a protein-protein by maintaining the identity of the cardiomyocyte. On the
interaction (PPI) network was created. Integrins (ITGA1, other hand, individuals with higher COL1A1 levels had
ITGB1), members of the collagen family (COL1A1, early unfavorable clinical outcomes; their median survival
COL3A1), and TGFB/SMAD signalling proteins was 2.5 years as opposed to 4.8 years (log-rank
Running Title Title of the Journal, Year, Vol. 0, No. 0 11

p=0.0035). Signs of an overabundance of fibroblasts groupings.

predicted a poorer course for heart failure.[48]
Patients were further classified by CDH5 levels, with The important and main findings of the study should com
greater percentage of endothelial cells indicating
protection (5.1 versus 3.2 years, log-rank p=0.047). In the
group with higher expression of the macrophage marker
CD68, there was a non-significant tendency towards
FIGURES, TABLES AND SCHEMES
worse results. After adjusting for age, aetiology, and
ejection percentage, multivariate Cox regression revealed Figures and Tables should be embedded in the text exactly
COL1A1 and CDH5 were independent predictors of according to their appropriate placement in the submitted
survival (COL1A1 HR 2.4, 95% CI 1.3-4.5, p=0.0048; manuscript.
CDH5 HR 0.62, 95% CI 0.41-0.94, p=0.025). These
findings show that, in addition to bulk pathogenic
characteristics, the cellular composition revealed by
single-cell profiling offers therapeutically significant
information. Fibroblasts and endothelial cells in particular
show promise as therapeutic targets due to their
correlation with long-term survival in heart failure.
Cellular signature stratification might improve medical
care.[49]
Validation Analysis:
Several internal and external techniques were used to
validate the heart failure subtypes that the convolutional
autoencoder network had discovered. By dividing the
dataset into 80% for training and 20% for testing, an
internal validation was carried out. The network
successfully classified the test samples that had not yet
been seen into the appropriate subtypes with an accuracy
of 85%, indicating the subtypes' resilience to fresh data.
The dataset was split into 5 folds, and the network was
trained and tested 5 times, holding out a separate fold for
validation, as part of a 5-fold cross-validation that was
also carried out. The subtypes generalize to various data
partitions, as evidenced by the mean classification
accuracy of 84% obtained across the folds. The subtypes
were also validated using consensus clustering, which
produced an agreement score of more than 85% and a Fig. (1). -
Venn Diagram for the upregulated and downregulated genes
consensus matrix, indicating the subtypes represent stable
clusters within the data.
The trained network was used to an independent dataset
of 150 patients with heart failure from a different medical
facility in order to carry out an external validation.
Verifying that the identified subtypes are applicable to a
patient population outside of the network, the network
accurately classified 70% of these additional external
samples based on known diagnoses. Additionally, the
subtypes showed notable variations in measures such as
heart size, comorbidities, and ejection fraction, which are
consistent with different heart failure etiologies,
according to further clinical validation. This affirmed the
unique pathophysiological characteristics of the identified
categories.
Convolutional autoencoder network on cardiac MRI
images yielded heart failure classes that were resilient and Fig. (2).
If a figure is in separate parts, all parts of the figure must be provided
broadly applicable, as demonstrated by internal validation
tests, cross-validation, consensus clustering, and external
clinical validation. The subtypes show how heart failure
has been steadily categorized into several, data-driven
12 Title of the Journal, Year, Vol. 0, No. 0 Principle Author et al.

The statement regarding the approval by an independent local, regi

Table 1.
Each table should include a title/caption being explanatory in itself with respect to the details discussed in the table. The title should be in smal

HUMAN AND ANIMAL RIGHTS

S. No. Title 1 Title 1 References Research Involving Humans
1. entry 1 Data [Mention
reference number] All clinical investigations should be conducted according to theDe

2. entry 2 Data - Research Involving Animals

3. entry 3 *Data -

* Symbols and nonstandard abbreviations should

for the Care and Use of Laboratory Animals”
be explained in the end of the text. For research involving animals, the authors should indicate whethe

All references should be numbered sequentially [in square brackets] in the table and listed in the same numerical order in the referenc

Research work on animals should be carried out in accordance wit

Columns and rows of data should be made visibly distinct by ensuring that the borders of each cell are displayed as black lines. Table

Research Involving Plants

5. DISCUSSION

This should explore the significance of the results of the work, present a reproducibleresearch
All experimental procedure and emphasis
on plants the importance
(either cultivated of the
or wild), shoa

CONSENT
The Results and Discussion may be presented together under one heading ofFOR PUBLICATION
“Results and Discussion”. Alternatively, they may be pr

If the manuscript has an individuals’ data, such as personal detail,

CONCLUSION

A small paragraph summarizing the contents of the article, presenting the final outcome of the research or proposing further study on

A specific declaration of such approval and consent-to-disclose for

LIST OF ABBREVIATIONS

If abbreviations are used in the text either they should be defined in the text where first used, or a list of abbreviations can be provide
AVAILABILITY OF DATA AND MATERIALS
ETHICS APPROVAL AND CONSENT TO
PARTICIPATE
The source of data and materials should be mentioned in the manu

Authors should clearly state the name of the approval committee, highlighting that legal and ethical approvals were obtained prior to
Running Title Title of the Journal, Year, Vol. 0, No. 0 13

The statement relating to the data should be presented in the following format under a separate ‘Availability of Data and Materials’ se
Financial contributions and any potential conflict of interest must b

ACKNOWLEDGEMENTS
"The data supporting the findings of the article is available in the [repository name] at [URL], reference number [reference number]”.

FUNDING
All individuals listed as authors must have contributed substantiall

The authors need to declare the funding sources of their manuscripts clearly by providing the name of the funding agency or financial

REFERENCES

References should be provided exactly in the journal's

specific format.
Similarly, if a paper does not have any specific funding source, and is part of the employment of the authors, then the name of the em
The references should be relevant to the study and should
refer to the original research sources. Self-citation and self-
CONFLICT OF INTEREST interest should be avoided by the authors, editors and peer-
reviewers.