DITHER: an algorithm for Defining IntraTumor Heterogeneity based on EntRopy

Pathway Enrichment-Based Unsupervised Learning Identifies Novel Subtypes of Cancer-Associated Fibroblasts in Pancreatic Ductal Adenocarcinoma

Existing single-cell clustering methods are based on gene expressions that are susceptible to dropout events in single-cell RNA sequencing (scRNA-seq) data. To overcome this limitation, we proposed a pathway-based clustering method for single cells (scPathClus). scPathClus first transforms the single-cell gene expression matrix into a pathway enrichment matrix and generates its latent feature matrix. Based on the latent feature matrix, scPathClus clusters single cells using the method of community detection. Applying scPathClus to pancreatic ductal adenocarcinoma (PDAC) scRNA-seq datasets, we identified two types of cancer-associated fibroblasts (CAFs), termed csCAFs and gapCAFs, which highly expressed complement system and gap junction-related pathways, respectively. Spatial transcriptome analysis revealed that gapCAFs and csCAFs are located at cancer and non-cancer regions, respectively. Pseudotime analysis suggested a potential differentiation trajectory from csCAFs to gapCAFs. Bulk transcriptome analysis showed that gapCAFs-enriched tumors are more endowed with tumor-promoting characteristics and worse clinical outcomes, while csCAFs-enriched tumors confront stronger antitumor immune responses. Compared to established CAF subtyping methods, this method displays better prognostic relevance.

Inferring tumor purity using multi-omics data based on a uniform machine learning framework MoTP

Existing algorithms for assessing tumor purity are limited to a single omics data, such as gene expression, somatic copy number variations, somatic mutations, and DNA methylation. Here we proposed the machine learning Multi-omics Tumor Purity prediction (MoTP) algorithm to estimate tumor purity based on multiple types of omics data. MoTP utilizes the Bayesian Regularized Neural Networks as the prediction algorithm, and Consensus Tumor Purity Estimates as labels. We trained MoTP using multi-omics data (mRNA, microRNA, long non-coding RNA, and DNA methylation) across 21 TCGA solid cancer types. By testing MoTP in TCGA validation sets, TCGA test sets, and eight datasets outside the TCGA cancer cohorts, we showed that although MoTP could achieve excellent performance in predicting tumor purity based on a single omics data type, the integration of multiple single omics data-based predictions can enhance the prediction performance. Moreover, we demonstrated the robustness of MoTP by testing it in datasets with Gaussian noise and feature missing. Benchmark analysis showed that MoTP outperformed most established tumor purity prediction algorithms, and that it required less running time and computational resource to fulfill the predictive task. Thus, MoTP would be an attractive option for computational tumor purity inference.

DEPTH2 score was associated with cell proliferation and immune cell infiltrations but not with systemic treatment response in breast cancer

Intratumoral genomic heterogeneity (ITGH), the existence of genotypic and phenotypic variation within an individual tumor, is known to be a key mechanism in treatment resistance. Deviating gene Expression Profiling Tumor Heterogeneity 2 (DEPTH2) algorithm was developed to estimate ITGH using solely RNA expression data unlike the others that require both DNA- and RNA-expression data. Total of 6,500 breast cancer patients from multiple independent cohorts were analyzed using DEPTH2. High DEPTH2 score patients were associated with worse overall survival consistently across all subtypes in METABRIC, but not in TCGA and SCAN-B cohort. Higher DEPTH2 score was linked to increased cell proliferation, as evidenced by elevated Nottingham histological grades and Ki67 gene expression, as well as enrichment of the cell proliferation-related gene sets, and immune cell infiltrations. DEPTH2 score was significantly higher in triple negative breast cancer among the subtypes but did not reflect with lymph node and distal metastasis. DEPTH2 scores decreased in two but showed no change in another two cohorts after neoadjuvant chemotherapy (NAC). DEPTH2 score was not associated with pathologic complete response after NAC in any subtypes across 3 cohorts. DEPTH2 score may not capture the entire biological aspects of ITGH in breast cancer patients.

The baseline hemoglobin level is a positive biomarker for immunotherapy response and can improve the predictability of tumor mutation burden for immunotherapy response in cancer

Purpose

Because only a subset of cancer patients can benefit from immunotherapy, identifying predictive biomarkers of ICI therapy response is of utmost importance.

Methods

We analyzed the association between hemoglobin (HGB) levels and clinical outcomes in 1,479 ICIs-treated patients across 16 cancer types. We explored the dose-dependent associations between HGB levels and survival and immunotherapy response using the spline-based cox regression analysis. Furthermore, we investigated the associations across subgroups of patients with different clinicopathological characteristics, treatment programs and cancer types using the bootstrap resampling method.

Results

HGB levels correlated positively with clinical outcomes in cancer patients receiving immunotherapy but not in those without immunotherapy. Moreover, this association was independent of other clinicopathological characteristics (such as sex, age, tumor stage and tumor mutation burden (TMB)), treatment program and cancer type. Also, this association was independent of the established biomarkers of immunotherapy response, including TMB, PD-L1 expression and microsatellite instability. The combination of TMB and HGB level are more powerful in predicting immunotherapy response than TMB alone. Multi-omics analysis showed that HGB levels correlated positively with antitumor immune signatures and negatively with tumor properties directing antitumor immunosuppression, such as homologous recombination defect, stemness and intratumor heterogeneity.

Conclusion

The HGB measure has the potential clinical value as a novel biomarker of immunotherapy response that is easily accessible from clinically routine examination. The combination of TMB and HGB measures have better predictive performance for immunotherapy response than TMB.

A comprehensive comparison of molecular and phenotypic profiles between hepatitis B virus (HBV)-infected and non-HBV-infected hepatocellular carcinoma by multi-omics analysis

Hepatitis B virus (HBV) infection is a major etiology of hepatocellular carcinoma (HCC). An interesting question is how different are the molecular and phenotypic profiles between HBV-infected (HBV+) and non-HBV-infected (HBV-) HCCs? Based on the publicly available multi-omics data for HCC, including bulk and single-cell data, and the data we collected and sequenced, we performed a comprehensive comparison of molecular and phenotypic features between HBV+ and HBV- HCCs. Our analysis showed that compared to HBV- HCCs, HBV+ HCCs had significantly better clinical outcomes, higher degree of genomic instability, higher enrichment of DNA repair and immune-related pathways, lower enrichment of stromal and oncogenic signaling pathways, and better response to immunotherapy. Furthermore, in vitro experiments confirmed that HBV+ HCCs had higher immunity, PD-L1 expression and activation of DNA damage response pathways. This study may provide insights into the profiles of HBV+ and HBV- HCCs, and guide rational therapeutic interventions for HCC patients.

Global analysis of iron metabolism-related genes identifies potential mechanisms of gliomagenesis and reveals novel targets

AIMS

This study aimed to investigate key regulators of aberrant iron metabolism in gliomas, and evaluate their effect on biological functions and clinical translational relevance.

METHODS

We used transcriptomic data from multiple cross-platform glioma cohorts to identify key iron metabolism-related genes (IMRGs) based on a series of bioinformatic and machine learning methods. The associations between IMRGs and prognosis, mesenchymal phenotype, and genomic alterations were analyzed in silico. The performance of the IMRGs-based signature in predicting temozolomide (TMZ) treatment sensitivity was evaluated. In vitro and in vivo experiments were used to explore the biological functions of these key IMRGs.

RESULTS

HMOX1, LTF, and STEAP3 were identified as the most essential IMRGs in gliomas. The expression levels of these genes were strongly related to clinicopathological and molecular features. The robust IMRG-based gene signature could be used for prognosis prediction. These genes facilitate mesenchymal transformation, driver gene mutations, and oncogenic alterations in gliomas. The gene signature was also associated with TMZ resistance. HMOX1, LTF, and STEAP3 knockdown in glioma cells significantly reduced cell proliferation, colony formation, migration, and malignant invasion.

CONCLUSION

The study presented a comprehensive view of key regulators underpinning iron metabolism in gliomas and provided new insights into novel therapeutic approaches.

Identifying biomarkers associated with immunotherapy response in melanoma by multi-omics analysis

Despite immune checkpoint inhibitors (ICIs) have shown the greatest success in melanoma treatment, only a subset of melanoma patients responds well to ICIs. Thus, identifying predictive biomarkers for immunotherapy response is crucial. In this study, we took complementary advantages of immunotherapy data and The Cancer Genome Atlas (TCGA) multi-omics data to explore the predictive biomarkers for the response to immunotherapy in melanoma. We first predicted responsive and non-responsive melanomas in the TCGA skin cutaneous melanoma (SKCM) cohort based on both somatic mutation and transcriptome datasets which involved immunotherapy data for melanoma. This method identified 170 responsive and 56 non-responsive melanomas in TCGA-SKCM. Based on the TCGA-SKCM data, we performed a comprehensive comparison of multi-omics molecular features between responsive and non-responsive melanomas. We identified the molecular features significantly associated with immunotherapy response in melanoma at the genome, transcriptome, epigenome, and proteome levels, respectively. Our analysis confirmed certain immunotherapy response-associated biomarkers, such as tumor mutation burden (TMB), copy number alteration (CNA), intratumor heterogeneity (ITH), PD-L1 expression, and tumor immunity. Moreover, we identified some novel molecular features associated with immunotherapy response: (1) the activation of mast cells and dendritic cells correlating negatively with immunotherapy response; (2) the enrichment of many oncogenic pathways correlating positively with immunotherapy response, such as JAK-STAT, RAS, MAPK, HIF-1, PI3K-Akt, and VEGF pathways; and (3) a number of microRNAs and proteins whose expression correlates with immunotherapy response. In addition, the mTOR signaling pathway has a negative association with immunotherapy response. The novel biomarkers have potential predictive values in immunotherapy response and warrant further investigation.

PreCanCell: An ensemble learning algorithm for predicting cancer and non-cancer cells from single-cell transcriptomes

We propose PreCanCell, a novel algorithm for predicting malignant and non-malignant cells from single-cell transcriptomes. PreCanCell first identifies the differentially expressed genes (DEGs) between malignant and non-malignant cells commonly in five common cancer types-associated single-cell transcriptome datasets. The five common cancer types include renal cell carcinoma (RCC), head and neck squamous cell carcinoma (HNSCC), melanoma, lung adenocarcinoma (LUAD), and breast cancer (BC). With each of the five datasets as the training set and the DEGs as the features, a single cell is classified as malignant or non-malignant by k-NN (k = 5). Finally, the single cell is determined as malignant or non-malignant by the majority vote of the five k-NN classification results. We tested the predictive performance of PreCanCell in 19 single-cell datasets, and reported classification accuracy, sensitivity, specificity, balanced accuracy (the average of sensitivity and specificity) and the area under the receiver operating characteristic curve (AUROC). In all these datasets, PreCanCell achieved above 0.8 accuracy, sensitivity, specificity, balanced accuracy and AUROC. Finally, we compared the predictive performance of PreCanCell with that of seven other algorithms, including CHETAH, SciBet, SCINA, scmap-cell, scmap-cluster, SingleR, and ikarus. Compared to these algorithms, PreCanCell displays the advantages of higher accuracy and simpler implementation. We have developed an R package for the PreCanCell algorithm, which is available at https://github.com/WangX-Lab/PreCanCell.

Gene set enrichment analysis identifies immune subtypes of kidney renal clear cell carcinoma with significantly different molecular and clinical properties

Background

Kidney renal clear cell carcinoma (KIRC) is the most prevalent renal malignancy, marked by a high abundance of tumor-infiltrating lymphocytes (TILs) and an unfavorable prognosis upon metastasis. Numerous studies have demonstrated that KIRC possesses a tumor microenvironment that is highly heterogeneous, and this is associated with significant variations in the effectiveness of most first-line drugs administered to KIRC patients. Therefore, it is crucial to classify KIRC based on the tumor microenvironment, although these subtyping techniques are still inadequate.

Methods

By applying gene set enrichment scores of 28 immune signatures, we conducted a hierarchical clustering of KIRC and determined its immune subtypes. In addition, we conducted a comprehensive exploration of the molecular and clinical features of these subtypes, including survival prognosis, proliferation, stemness, angiogenesis, tumor microenvironment, genome instability, intratumor heterogeneity, and pathway enrichment.

Results

Through cluster analysis, two immune subtypes of KIRC were identified and termed Immunity-High (Immunity-H) and Immunity-Low (Immunity-L). This clustering outcome was consistent in four independent KIRC cohorts. The subtype Immunity-H exhibited elevated levels of TILs, tumor aneuploidy, homologous recombination deficiency, stemness, and proliferation potential, along with a poorer prognosis for survival. Despite this, the Immunity-L subtype demonstrated elevated intratumor heterogeneity and a stronger angiogenesis signature in contrast to Immunity-H. According to the results of pathway enrichment analysis, the Immunity-H subtype was found to be highly enriched in immunological, oncogenic, and metabolic pathways, whereas the Immunity-L subtype was highly enriched in angiogenic, neuroactive ligand-receptor interaction, and PPAR pathways.

Conclusions

Based on the enrichment of immune signatures in the tumor microenvironment, KIRC can be categorized into two immune subtypes. The two subtypes demonstrate considerably distinct molecular and clinical features. In KIRC, an increase in immune infiltration is linked to a poor prognosis. Patients with Immunity-H KIRC may exhibit active responses to PPAR and immune checkpoint inhibitors, whereas patients with Immunity-L may manifest favorable responses to anti-angiogenic agents and immune checkpoint inhibitors. The immunological classification provides molecular insights into KIRC immunity, as well as clinical implications for the management of this disease.

Defining multiple layers of intratumor heterogeneity based on variations of perturbations in multi-omics profiling

BACKGROUND

Intratumor heterogeneity (ITH) plays a crucial role in tumor progression, relapse, immune evasion, and drug resistance. Existing ITH quantification methods based on a single molecular level are inadequate to capture ITH evolving from genotype to phenotype.

METHODS

We designed a set of information entropy (IE)-based algorithms for quantifying ITH at the genome (somatic copy number alterations and mutations), mRNA, microRNA (miRNA), long non-coding RNA (lncRNA), protein, and epigenome level, respectively. We evaluated the performance of these algorithms by analyzing the correlations between their ITH scores and ITH-associated molecular and clinical features in 33 TCGA cancer types. Moreover, we evaluated the correlations between the ITH measures at different molecular levels by Spearman correlation and clustering analysis.

RESULTS

The IE-based ITH measures had significant correlations with unfavorable prognosis, tumor progression, genomic instability, antitumor immunosuppression, and drug resistance. The mRNA ITH showed stronger correlations with the miRNA, lncRNA, and epigenome ITH than with the genome ITH, supporting the regulatory relationships of miRNA, lncRNA, and DNA methylation towards mRNA. The protein-level ITH displayed stronger correlations with the transcriptome-level ITH than with the genome-level ITH, supporting the central dogma of molecular biology. Clustering analysis based on the ITH scores identified four subtypes of pan-cancer showing significantly different prognosis. Finally, the ITH integrating the seven ITH measures displayed more prominent properties of ITH than that at a single level.

CONCLUSIONS

This analysis provides landscapes of ITH at various molecular levels. Combining the ITH observation from different molecule levels will improve personalized management for cancer patients.

Identifying Immune-Specific Subtypes of Adrenocortical Carcinoma Based on Immunogenomic Profiling

BACKGROUND

The tumor immune microenvironment (TIME) of adrenocortical carcinoma (ACC) is heterogeneous. However, a classification of ACC based on the TIME remains unexplored.

METHODS

We hierarchically clustered ACC based on the enrichment levels of twenty-three immune signatures to identify its immune-specific subtypes. Furthermore, we comprehensively compared the clinical and molecular profiles between the subtypes.

RESULTS

We identified two immune-specific subtypes of ACC: Immunity-H and Immunity-L, which had high and low immune signature scores, respectively. We demonstrated that this subtyping method was stable and reproducible by analyzing five different ACC cohorts. Compared with Immunity-H, Immunity-L had lower levels of immune cell infiltration, worse overall and disease-free survival prognosis, and higher tumor stemness, genomic instability, proliferation potential, and intratumor heterogeneity. Furthermore, the ACC driver gene CTNNB1 was more frequently mutated in Immunity-L than in Immunity-H. Several proteins, such as mTOR, ERCC1, Akt, ACC1, Cyclin_E1, β-catenin, FASN, and GAPDH, were more highly expressed in Immunity-L than in Immunity-H. In contrast, p53, Syk, Lck, PREX1, and MAPK were more highly expressed in Immunity-H. Pathway and gene ontology analysis showed that the immune, stromal, and apoptosis pathways were highly enriched in Immunity-H, while the cell cycle, steroid biosynthesis, and DNA damage repair pathways were highly enriched in Immunity-L.

CONCLUSIONS

ACC can be classified into two stable immune-related subtypes, which have significantly different antitumor responses, molecular characteristics, and clinical outcomes. This subtyping may provide clinical implications for prognostic and immunotherapeutic stratification of ACC.

Immunological classification of hepatitis B virus-positive hepatocellular carcinoma by transcriptome analysis

BACKGROUND

Hepatitis B virus (HBV) infection is a major factor responsible for HBV+ hepatocellular carcinoma (HCC).

AIM

An immunological classification of HBV+ HCC may provide both biological insights and clinical implications for this disease.

METHODS

Based on the enrichment of 23 immune signatures, we identified two immune-specific subtypes (Imm-H and Imm-L) of HBV+ HCC by unsupervised clustering. We showed that this subtyping method was reproducible and predictable by analyzing three different datasets.

RESULTS

Compared to Imm-L, Imm-H displayed stronger immunity, more stromal components, lower tumor purity, lower stemness and intratumor heterogeneity, lower-level copy number alterations, higher global methylation level, and better overall and disease-free survival prognosis. Besides immune-related pathways, stromal pathways (ECM receptor interaction, focal adhesion, and regulation of actin cytoskeleton) and neuro-related pathways (neuroactive ligand-receptor interaction, and prion diseases) were more highly enriched in Imm-H than in Imm-L. We identified nine proteins differentially expressed between Imm-H and Imm-L, of which MYH11, PDCD4, Dvl3, and Syk were upregulated in Imm-H, while PCNA, Acetyl-a-Tubulin-Lys40, ER-α_pS118, Cyclin E2, and β-Catenin were upregulated in Imm-L.

CONCLUSION

Our data suggest that “hot” tumors have a better prognosis than “cold” tumors in HBV+ HCC and that “hot” tumors respond better to immunotherapy.

TRPV1 Is a Potential Tumor Suppressor for Its Negative Association with Tumor Proliferation and Positive Association with Antitumor Immune Responses in Pan-Cancer

Background

Although numerous studies have shown that the expression and activation of TRPV1 have an important role in cancer development, a comprehensive exploration of associations between TRPV1 expression and tumor proliferation, microenvironment, and clinical outcomes in pan-cancer remains insufficient.

Methods

From The Cancer Genome Atlas (TCGA) program, we downloaded multiomics data of ten cancer cohorts and investigated the correlations between TRPV1 expression and immune signatures' enrichment, stromal content, genomic features, oncogenic signaling, and clinical features in these cancer cohorts and pan-cancer.

Results

Elevated expression of TRPV1 correlated with better clinical outcomes in pan-cancer and diverse cancer types. In multiple cancer types, TRPV1 expression correlated negatively with the expression of tumor proliferation marker genes (MKI67 and RACGAP1), proliferation scores, cell cycle scores, stemness scores, epithelial-mesenchymal transition scores, oncogenic pathways' enrichment, tumor immunosuppressive signals, intratumor heterogeneity, homologous recombination deficiency, tumor mutation burden, and stromal content. Moreover, TRPV1 expression was downregulated in late-stage versus early-stage tumors. In breast cancer, bladder cancer, and low-grade glioma, TRPV1 expression was more inferior in invasive than in noninvasive subtypes. Pathway analysis showed that the enrichment of cancer-associated pathways correlated inversely with TRPV1 expression levels.

Conclusion

TRPV1 upregulation correlates with decreased tumor proliferation, tumor driver gene expression, genomic instability, and tumor immunosuppressive signals in various cancers. Our results provide new understanding of the role of TRPV1 in both cancer biology and clinical practice.

Molecular classification of human papillomavirus-positive cervical cancers based on immune signature enrichment

Background

Human papillomavirus-positive (HPV+) cervical cancers are highly heterogeneous in clinical and molecular characteristics. Thus, an investigation into their heterogeneous immunological profiles is meaningful in providing both biological and clinical insights into this disease.

Methods

Based on the enrichment of 29 immune signatures, we discovered immune subtypes of HPV+ cervical cancers by hierarchical clustering. To explore whether this subtyping method is reproducible, we analyzed three bulk and one single cell transcriptomic datasets. We also compared clinical and molecular characteristics between the immune subtypes.

Results

Clustering analysis identified two immune subtypes of HPV+ cervical cancers: Immunity-H and Immunity-L, consistent in the four datasets. In comparisons with Immunity-L, Immunity-H displayed stronger immunity, more stromal contents, lower tumor purity, proliferation potential, intratumor heterogeneity and stemness, higher tumor mutation burden, more neoantigens, lower levels of copy number alterations, lower DNA repair activity, as well as better overall survival prognosis. Certain genes, such as MUC17, PCLO, and GOLGB1, showed significantly higher mutation rates in Immunity-L than in Immunity-H. 16 proteins were significantly upregulated in Immunity-H vs. Immunity-L, including Caspase-7, PREX1, Lck, C-Raf, PI3K-p85, Syk, 14-3-3_epsilon, STAT5-α, GATA3, Src_pY416, NDRG1_pT346, Notch1, PDK1_pS241, Bim, NF-kB-p65_pS536, and p53. Pathway analysis identified numerous immune-related pathways more highly enriched in Immunity-H vs. Immunity-L, including cytokine-cytokine receptor interaction, natural killer cell-mediated cytotoxicity, antigen processing and presentation, T/B cell receptor signaling, chemokine signaling, supporting the stronger antitumor immunity in Immunity-H vs. Immunity-L.

Conclusion

HPV+ cervical cancers are divided into two subgroups based on their immune signatures' enrichment. Both subgroups have markedly different tumor immunity, progression phenotypes, genomic features, and clinical outcomes. Our data offer novel perception in the tumor biology as well as clinical implications for HPV+ cervical cancer.

Dissection of Immune Profiles in Microsatellite Stable and Low Microsatellite Instability Colon Adenocarcinoma by Multiomics Data Analysis

Background

Although microsatellite instability (MSI) is an indicator for active immunotherapy response, only 15% of colon adenocarcinoma (COAD) patients are with MSI. An investigation into the immune profiles in low MSI (MSI-L) and microsatellite stable (MSS) COAD remains lacking, whereas such exploration may provide new insights into COAD immunity.

Methods

We hierarchically clustered MSI-L/MSS COAD based on the enrichment levels of 28 immune signatures to identify its immune-specific subtypes. We also comprehensively compared molecular and clinicopathologic profiles among these subtypes.

Results

We identified three immune subtypes of MSI-L/MSS COAD (IM-H, IM-M, and IM-L), which had high, medium, and low immune signature scores, respectively. We demonstrated that this subtyping method was reproducible and predictable by analyzing five different datasets, including four bulk tumor datasets and one single-cell dataset. IM-H was characterized by high immunity, high stemness, strong potential of proliferation, invasion and metastasis, epithelial-mesenchymal transition, elevated expression of oncogenic pathways, low tumor purity, low intratumor heterogeneity (ITH), genomic instability, inferior response to chemotherapy, and unfavorable prognosis. IM-M was characterized by the highest ratio of immunostimulatory to immunosuppressive signatures, the best response to chemotherapy, and favorable prognosis. IM-L was characterized by low immunity, high tumor purity, high ITH, and genomic stability.

Conclusion

The immune-specific subtyping of MSI-L/MSS COAD may provide new insights into the tumor immunity as well as clinical implications for immunotherapy of the COAD patients who lack MSI.

Subtyping of Human Papillomavirus-Positive Cervical Cancers Based on the Expression Profiles of 50 Genes

Background

Human papillomavirus-positive (HPV+) cervical cancers are highly heterogeneous in molecular and clinical features. However, the molecular classification of HPV+ cervical cancers remains insufficiently unexplored.

Methods

Based on the expression profiles of 50 genes having the largest expression variations across the HPV+ cervical cancers in the TCGA-CESC dataset, we hierarchically clustered HPV+ cervical cancers to identify new subtypes. We further characterized molecular, phenotypic, and clinical features of these subtypes.

Results

We identified two subtypes of HPV+ cervical cancers, namely HPV+G1 and HPV+G2. We demonstrated that this classification method was reproducible in two validation sets. Compared to HPV+G2, HPV+G1 displayed significantly higher immune infiltration level and stromal content, lower tumor purity, lower stemness scores and intratumor heterogeneity (ITH) scores, higher level of genomic instability, lower DNA methylation level, as well as better disease-free survival prognosis. The multivariate survival analysis suggests that the disease-free survival difference between both subtypes is independent of confounding variables, such as immune signature, stemness, and ITH. Pathway and gene ontology analysis confirmed the more active tumor immune microenvironment in HPV+G1 versus HPV+G2.

Conclusions

HPV+ cervical cancers can be classified into two subtypes based on the expression profiles of the 50 genes with the largest expression variations across the HPV+ cervical cancers. Both subtypes have significantly different molecular, phenotypic, and clinical features. This new subtyping method captures the comprehensive heterogeneity in molecular and clinical characteristics of HPV+ cervical cancers and provides potential clinical implications for the diagnosis and treatment of this disease.