A Machine Learning Model to Predict the Triple Negative Breast Cancer Immune Subtype

Tumor purity-related genes for predicting the prognosis and drug sensitivity of DLBCL patients

Background

Diffuse large B-cell lymphoma (DLBCL) is the predominant type of malignant B-cell lymphoma. Although various treatments have been developed, the limited efficacy calls for more and further exploration of its characteristics.

Methods

Datasets from the Gene Expression Omnibus (GEO) database were used for identifying the tumor purity of DLBCL. Survival analysis was employed for analyzing the prognosis of DLBCL patients. Immunohistochemistry was conducted to detect the important factors that influenced the prognosis. Drug-sensitive prediction was performed to evaluate the value of the model.

Results

VCAN, CD3G, and C1QB were identified as three key genes that impacted the outcome of DLBCL patients both in GEO datasets and samples from our center. Among them, VCAN and CD3G+ T cells were correlated with favorable prognosis, and C1QB was correlated with worse prognosis. The ratio of CD68 + macrophages and CD8 + T cells was associated with better prognosis. In addition, CD3G+T cells ratio was significantly correlated with CD68 + macrophages, CD4 + T cells, and CD8 +T cells ratio, indicating it could play an important role in the anti-tumor immunity in DLBCL. The riskScore model constructed based on the RNASeq data of VCAN, C1QB, and CD3G work well in predicting the prognosis and drug sensitivity.

Conclusions

VCAN, CD3G, and C1QB were three key genes that influenced the tumor purity of DLBCL, and could also exert certain impact on drug sensitivity and prognosis of DLBCL patients.

Funding

This work is supported by the Shenzhen High-level Hospital Construction Fund and CAMS Innovation Fund for Medical Sciences (CIFMS) (2022-I2M-C&T-B-062).

A novel transformer-based aggregation model for predicting gene mutations in lung adenocarcinoma

In recent years, predicting gene mutations on whole slide imaging (WSI) has gained prominence. The primary challenge is extracting global information and achieving unbiased semantic aggregation. To address this challenge, we propose a novel Transformer-based aggregation model, employing a self-learning weight aggregation mechanism to mitigate semantic bias caused by the abundance of features in WSI. Additionally, we adopt a random patch training method, which enhances model learning richness by randomly extracting feature vectors from WSI, thus addressing the issue of limited data. To demonstrate the model's effectiveness in predicting gene mutations, we leverage the lung adenocarcinoma dataset from Shandong Provincial Hospital for prior knowledge learning. Subsequently, we assess TP53, CSMD3, LRP1B, and TTN gene mutations using lung adenocarcinoma tissue pathology images and clinical data from The Cancer Genome Atlas (TCGA). The results indicate a notable increase in the AUC (Area Under the ROC Curve) value, averaging 4%, attesting to the model's performance improvement. Our research offers an efficient model to explore the correlation between pathological image features and molecular characteristics in lung adenocarcinoma patients. This model introduces a novel approach to clinical genetic testing, expected to enhance the efficiency of identifying molecular features and genetic testing in lung adenocarcinoma patients, ultimately providing more accurate and reliable results for related studies.

Predicting prostate cancer recurrence: Introducing PCRPS, an advanced online web server

Background

Prostate cancer (PCa) is one of the leading causes of cancer death in men. About 30% of PCa will develop a biochemical recurrence (BCR) following initial treatment, which significantly contributes to prostate cancer-related deaths. In clinical practice, accurate prediction of PCa recurrence is crucial for making informed treatment decisions. However, the development of reliable models and biomarkers for predicting PCa recurrence remains a challenge. In this study, the aim is to establish an effective and reliable tool for predicting the recurrence of PCa.

Methods

We systematically screened and analyzed potential datasets to predict PCa recurrence. Through quality control analysis, low-quality datasets were removed. Using meta-analysis, differential expression analysis, and feature selection, we identified key genes associated with recurrence. We also evaluated 22 previously published signatures for PCa recurrence prediction. To assess prediction performance, we employed nine machine learning algorithms. We compared the predictive capabilities of models constructed using clinical variables, expression data, and their combinations. Subsequently, we implemented these machine learning models into a user-friendly web server freely accessible to all researchers.

Results

Based on transcriptomic data derived from eight multicenter studies consisting of 733 PCa patients, we screened 23 highly influential genes for predicting prostate cancer recurrence. These genes were used to construct the Prostate Cancer Recurrence Prediction Signature (PCRPS). By comparing with 22 published signatures and four important clinicopathological features, the PCRPS exhibited a robust and significantly improved predictive capability. Among the tested algorithms, Random Forest demonstrated the highest AUC value of 0.72 in predicting PCa recurrence in the testing dataset. To facilitate access and usage of these machine learning models by all researchers and clinicians, we also developed an online web server (https://urology1926.shinyapps.io/PCRPS/) where the PCRPS model can be freely utilized. The tool can also be used to (1) predict the PCa recurrence by clinical information or expression data with high accuracy. (2) provide the possibility of PCa recurrence by nine machine learning algorithms. Furthermore, using the PCRPS scores, we predicted the sensitivity of 22 drugs from GDSC2 and 95 drugs from CTRP2 to the samples. These predictions provide valuable insights into potential drug sensitivities related to the PCRPS score groups.

Conclusion

Overall, our study provides an attractive tool to further guide the clinical management and individualized treatment for PCa.

Identification of hub genes and diagnostic efficacy for triple-negative breast cancer through WGCNA and Mendelian randomization

OBJECTIVE

Triple-negative breast cancer (TNBC) represents a particularly aggressive form of breast cancer with a poor prognosis due to a lack of targeted treatments resulting from limited a understanding of the underlying mechanisms. The aim of this study was the identification of hub genes for TNBC and assess their clinical applicability in predicting the disease.

METHODS

This study employed a combination of weighted gene co-expression network analysis (WGCNA) and differentially expressed genes (DEGs) to identify new susceptible modules and central genes in TNBC. The potential functional roles of the central genes were investigated using Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses. Furthermore, a predictive model and ROC curve were developed to assess the diagnostic performance of the identified central genes. The correlation between CCNB1 and immune cells proportion was also investigated. At last, a Mendelian randomization (MR) analysis utilizing Genome-Wide Association Study (GWAS) data was analyzed to establish the causal effect of CCNB1 level on TNBC.

RESULTS

WGCNA was applied to determine gene co-expression maps and identify the most relevant module. Through a screening process, 1585 candidate hub genes were subsequently identified with WGCNA and DEGs. GO and KEGG function enrichment analysis indicated that these core genes were related to various biological processes, such as organelle fission, chromosome segregation, nuclear division, mitotic cell cycle phase transition, the cell cycle, amyotrophic lateral sclerosis, and motor proteins. Using STRING and Cytoscape, the top five genes with high degrees were identified as CDC2, CCNB1, CCNA2, TOP2A, and CCNB2. The nomogram model demonstrated good performance in predicting TNBC risk and was proven effective in diagnosis, as evidenced by the receiver operating characteristic (ROC) curve. Further investigation revealed a causal association between CCNB1 and immune cell infiltrates in TNBC. Survival analysis revealed high expression of the CCNB1 gene leads to poorer prognosis in TNBC patients. Additionally, analysis using inverse variance weighting revealed that CCNB1 was linked to a 2.8% higher risk of TNBC (OR: 1.028, 95% CI 1.002-1.055, p = 0.032).

CONCLUSION

We established a co-expression network using the WGCNA methodology to detect pivotal genes associated with TNBC. This finding holds promise for advancing the creation of pre-symptomatic diagnostic tools and deepening our comprehension of the pathogenic mechanisms involved in TNBC risk genes.

Development and verification of a novel immunogenic cell death-related signature for predicting the prognosis and immune infiltration in triple-negative breast cancer

BACKGROUND

Insufficient understanding of the pathogenesis and tumor immunology of triple-negative breast cancer (TNBC) has limited the development of immunotherapy. The importance of tumor microenvironment (TME) in immunotyping, prognostic assessment and immunotherapy efficacy of cancer has been emphasized, however, potential immunogenic cell death (ICD) related genes function in TME of TNBC has been rarely investigated.

AIMS

To initially explore the role and related mechanisms of ICD in TNBC, especially the role played in the TME of TNBC, and to identify different relevant subtypes based on ICD, and then develop an ICD-related risk score to predict each TNBC patient TME status, prognosis and immunotherapy response.

METHODS AND RESULTS

In this study, we identified distinct ICD-related modification patterns based on 158 TNBC cases in the TCGA-TNBC cohort. We then investigated the possible correlation between ICD-related modification patterns and TME cell infiltration characteristics in TNBC. By using univariate Cox and least absolute shrinkage and selection operator (LASSO) regression analysis, we created a risk scoring system (ICD score) to quantifiably evaluate the impact of ICD-related modification patterns in individual TNBC patient. Two different ICD-related modification patterns were found with significant differences in immune infiltration. Lower ICD score was correlated with higher immune infiltration, tumor mutational burden and significantly enriched in immune-related pathways, indicating a strong ability to activate immune response, which might account for relatively favorable prognosis of TNBC patients and could serve as a predictor to select suitable candidates for immunotherapy. We used two independent cohorts, GSE58812 cohort and Metabric cohort to validate prognosis and immunohistochemistry for preliminary in vitro validation.

CONCLUSION

This study evidenced that the ICD-related modification patterns might exert pivotal roles in the immune infiltration landscape of TNBC and ICD score might act as potential predictors of prognostic assessment and immunotherapy response. This research provides unique insights for individualize immune treatment strategies and promising immunotherapy candidates screening.

NORAD-Regulated Signaling Pathways in Breast Cancer Progression

Long non-coding RNA activated by DNA damage (NORAD) has recently been associated with pathologic mechanisms underlying cancer progression. Due to NORAD's extended range of interacting partners, there has been contradictory data on its oncogenic or tumor suppressor roles in BC. This review will summarize the function of NORAD in different BC subtypes and how NORAD impacts crucial signaling pathways in this pathology. Through the preferential binding to pumilio (PUM) proteins PUM1 and PUM2, NORAD has been shown to be involved in the control of cell cycle, angiogenesis, mitosis, DNA replication and transcription and protein translation. More recently, NORAD has been associated with PUM-independent roles, accomplished by interacting with other ncRNAs, mRNAs and proteins. The intricate network of NORAD-mediated signaling pathways may provide insights into the potential design of novel unexplored strategies to overcome chemotherapy resistance in BC treatment.

Patient-specific signaling signatures predict optimal therapeutic combinations for triple negative breast cancer

Triple negative breast cancer (TNBC) is a heterogeneous group of tumors which lack estrogen receptor, progesterone receptor, and HER2 expression. Targeted therapies have limited success in treating TNBC, thus a strategy enabling effective targeted combinations is an unmet need. To tackle these challenges and discover individualized targeted combination therapies for TNBC, we integrated phosphoproteomic analysis of altered signaling networks with patient-specific signaling signature (PaSSS) analysis using an information-theoretic, thermodynamic-based approach. Using this method on a large number of TNBC patient-derived tumors (PDX), we were able to thoroughly characterize each PDX by computing a patient-specific set of unbalanced signaling processes and assigning a personalized therapy based on them. We discovered that each tumor has an average of two separate processes, and that, consistent with prior research, EGFR is a major core target in at least one of them in half of the tumors analyzed. However, anti-EGFR monotherapies were predicted to be ineffective, thus we developed personalized combination treatments based on PaSSS. These were predicted to induce anti-EGFR responses or to be used to develop an alternative therapy if EGFR was not present.In-vivo experimental validation of the predicted therapy showed that PaSSS predictions were more accurate than other therapies. Thus, we suggest that a detailed identification of molecular imbalances is necessary to tailor therapy for each TNBC. In summary, we propose a new strategy to design personalized therapy for TNBC using pY proteomics and PaSSS analysis. This method can be applied to different cancer types to improve response to the biomarker-based treatment.

Combining bioinformatics and machine learning algorithms to identify and analyze shared biomarkers and pathways in COVID-19 convalescence and diabetes mellitus

Background

Most patients who had coronavirus disease 2019 (COVID-19) fully recovered, but many others experienced acute sequelae or persistent symptoms. It is possible that acute COVID-19 recovery is just the beginning of a chronic condition. Even after COVID-19 recovery, it may lead to the exacerbation of hyperglycemia process or a new onset of diabetes mellitus (DM). In this study, we used a combination of bioinformatics and machine learning algorithms to investigate shared pathways and biomarkers in DM and COVID-19 convalescence.

Methods

Gene transcriptome datasets of COVID-19 convalescence and diabetes mellitus from Gene Expression Omnibus (GEO) were integrated using bioinformatics methods and differentially expressed genes (DEGs) were found using the R programme. These genes were also subjected to Gene Ontology (GO) functional enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis to find potential pathways. The hub DEGs genes were then identified by combining protein-protein interaction (PPI) networks and machine learning algorithms. And transcription factors (TFs) and miRNAs were predicted for DM after COVID-19 convalescence. In addition, the inflammatory and immune status of diabetes after COVID-19 convalescence was assessed by single-sample gene set enrichment analysis (ssGSEA).

Results

In this study, we developed genetic diagnostic models for 6 core DEGs beteen type 1 DM (T1DM) and COVID-19 convalescence and 2 core DEGs between type 2 DM (T2DM) and COVID-19 convalescence and demonstrated statistically significant differences (p<0.05) and diagnostic validity in the validation set. Analysis of immune cell infiltration suggests that a variety of immune cells may be involved in the development of DM after COVID-19 convalescence.

Conclusion

We identified a genetic diagnostic model for COVID-19 convalescence and DM containing 8 core DEGs and constructed a nomogram for the diagnosis of COVID-19 convalescence DM.

Machine learning-based models for the prediction of breast cancer recurrence risk

Breast cancer is the most common malignancy diagnosed in women worldwide. The prevalence and incidence of breast cancer is increasing every year; therefore, early diagnosis along with suitable relapse detection is an important strategy for prognosis improvement. This study aimed to compare different machine algorithms to select the best model for predicting breast cancer recurrence. The prediction model was developed by using eleven different machine learning (ML) algorithms, including logistic regression (LR), random forest (RF), support vector classification (SVC), extreme gradient boosting (XGBoost), gradient boosting decision tree (GBDT), decision tree, multilayer perceptron (MLP), linear discriminant analysis (LDA), adaptive boosting (AdaBoost), Gaussian naive Bayes (GaussianNB), and light gradient boosting machine (LightGBM), to predict breast cancer recurrence. The area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 score were used to evaluate the performance of the prognostic model. Based on performance, the optimal ML was selected, and feature importance was ranked by Shapley Additive Explanation (SHAP) values. Compared to the other 10 algorithms, the results showed that the AdaBoost algorithm had the best prediction performance for successfully predicting breast cancer recurrence and was adopted in the establishment of the prediction model. Moreover, CA125, CEA, Fbg, and tumor diameter were found to be the most important features in our dataset to predict breast cancer recurrence. More importantly, our study is the first to use the SHAP method to improve the interpretability of clinicians to predict the recurrence model of breast cancer based on the AdaBoost algorithm. The AdaBoost algorithm offers a clinical decision support model and successfully identifies the recurrence of breast cancer.

Identification and validation of novel biomarkers associated with immune infiltration for the diagnosis of osteosarcoma based on machine learning

Objectives: Osteosarcoma is the most common primary malignant tumor in children and adolescents, and the 5-year survival of osteosarcoma patients gained no substantial improvement over the past decades. Effective biomarkers in diagnosing osteosarcoma are warranted to be developed. This study aims to explore novel biomarkers correlated with immune cell infiltration in the development and diagnosis of osteosarcoma. Methods: Three datasets (GSE19276, GSE36001, GSE126209) comprising osteosarcoma samples were extracted from Gene Expression Omnibus (GEO) database and merged to obtain the gene expression. Then, differentially expressed genes (DEGs) were identified by limma and potential biological functions and downstream pathways enrichment analysis of DEGs was performed. The machine learning algorithms LASSO regression model and SVM-RFE (support vector machine-recursive feature elimination) analysis were employed to identify candidate hub genes for diagnosing patients with osteosarcoma. Receiver operating characteristic (ROC) curves were developed to evaluate the discriminatory abilities of these candidates in both training and test sets. Furthermore, the characteristics of immune cell infiltration in osteosarcoma, and the correlations between these potential genes and immune cell abundance were illustrated using CIBERSORT. qRT-PCR and western blots were conducted to validate the expression of diagnostic candidates. Results: GEO datasets were divided into the training (merged GSE19276, GSE36001) and test (GSE126209) groups. A total of 71 DEGs were screened out in the training set, including 10 upregulated genes and 61 downregulated genes. These DEGs were primarily enriched in immune-related biological functions and signaling pathways. After machine learning by SVM-RFE and LASSO regression model, four biomarkers were chosen for the diagnostic nomogram for osteosarcoma, including ASNS, CD70, SRGN, and TRIB3. These diagnostic biomarkers all possessed high diagnostic values (AUC ranging from 0.900 to 0.955). Furthermore, these genes were significantly correlated with the infiltration of several immune cells, such as monocytes, macrophages M0, and neutrophils. Conclusion: Four immune-related candidate hub genes (ASNS, CD70, SRGN, TRIB3) with high diagnostic value were confirmed for osteosarcoma patients. These diagnostic genes were significantly connected with the immune cell abundance, suggesting their critical roles in the osteosarcoma tumor immune microenvironment. Our study provides highlights on novel diagnostic candidate genes with high accuracy for diagnosing osteosarcoma patients.

Machine learning was used to predict risk factors for distant metastasis of pancreatic cancer and prognosis analysis

BACKGROUND

The mechanisms of distant metastasis in pancreatic cancer (PC) have not been elucidated, and this study aimed to explore the risk factors affecting the metastasis and prognosis of metastatic patients and to develop a predictive model.

METHOD

Clinical data from patients meeting criteria from 1990 to 2019 were obtained from the Surveillance, Epidemiology, and End Results (SEER) database, and two machine learning methods, random forest and support vector machine, combined with logistic regression, were used to explore risk factors influencing distant metastasis and to create nomograms. The performance of the model was validated using calibration curves and ROC curves based on the Shaanxi Provincial People's Hospital cohort. LASSO regression and Cox regression models were used to explore the independent risk factors affecting the prognosis of patients with distant PC metastases.

RESULTS

We found that independent risk factors affecting PC distant metastasis were: age, radiotherapy, chemotherapy, T and N; the independent risk factors for patient prognosis were: age, grade, bone metastasis, brain metastasis, lung metastasis, radiotherapy and chemotherapy.

CONCLUSION

Together, our study provides a method for risk factors and prognostic assessment for patients with distant PC metastases. The nomogram we developed can be used as a convenient individualized tool to facilitate aid in clinical decision making.

Application of Novel Transcription Factor Machine Learning Model and Targeted Drug Combination Therapy Strategy in Triple Negative Breast Cancer

Transcription factors (TFs) have been shown to play a key role in the occurrence and development of tumors, including triple-negative breast cancer (TNBC), with a worse prognosis. Machine learning is widely used for establishing prediction models and screening key tumor drivers. Current studies lack TF integration in TNBC, so targeted research on TF prognostic models and targeted drugs is beneficial to improve clinical translational application. The purpose of this study was to use the Least Absolute Shrinkage and Selection Operator to build a prognostic TFs model after cohort normalization based on housekeeping gene expression levels. Potential targeted drugs were then screened on the basis of molecular docking, and a multi-drug combination strategy was used for both in vivo and in vitro experimental studies. The machine learning model of TFs built by E2F8, FOXM1, and MYBL2 has broad applicability, with an AUC value of up to 0.877 at one year. As a high-risk clinical factor, its abnormal disorder may lead to upregulation of the activity of pathways related to cell proliferation. This model can also be used to predict the adverse effects of immunotherapy in patients with TNBC. Molecular docking was used to screen three drugs that target TFs: Trichostatin A (TSA), Doxorubicin (DOX), and Calcitriol. In vitro and in vivo experiments showed that TSA + DOX was able to effectively reduce DOX dosage, and TSA + DOX + Calcitriol may be able to effectively reduce the toxic side effects of DOX on the heart. In conclusion, the machine learning model based on three TFs provides new biomarkers for clinical and prognostic diagnosis of TNBC, and the combination targeted drug strategy offers a novel research perspective for TNBC treatment.

Prediction of the risk of cytopenia in hospitalized HIV/AIDS patients using machine learning methods based on electronic medical records

Background

Cytopenia is a frequent complication among HIV-infected patients who require hospitalization. It can have a negative impact on the treatment outcomes for these patients. However, by leveraging machine learning techniques and electronic medical records, a predictive model can be developed to evaluate the risk of cytopenia during hospitalization in HIV patients. Such a model is crucial for designing a more individualized and evidence-based treatment strategy for HIV patients.

Method

The present study was conducted on HIV patients who were admitted to Guangxi Chest Hospital between June 2016 and October 2021. We extracted a total of 66 clinical features from the electronic medical records and employed them to train five machine learning prediction models (artificial neural network [ANN], adaptive boosting [AdaBoost], k-nearest neighbour [KNN] and support vector machine [SVM], decision tree [DT]). The models were tested using 20% of the data. The performance of the models was evaluated using indicators such as the area under the receiver operating characteristic curve (AUC). The best predictive models were interpreted using the shapley additive explanation (SHAP).

Result

The ANN models have better predictive power. According to the SHAP interpretation of the ANN model, hypoproteinemia and cancer were the most important predictive features of cytopenia in HIV hospitalized patients. Meanwhile, the lower hemoglobin-to-RDW ratio (HGB/RDW), low-density lipoprotein cholesterol (LDL-C) levels, CD4+ T cell counts, and creatinine clearance (Ccr) levels increase the risk of cytopenia in HIV hospitalized patients.

Conclusion

The present study constructed a risk prediction model for cytopenia in HIV patients during hospitalization with machine learning and electronic medical record information. The prediction model is important for the rational management of HIV hospitalized patients and the personalized treatment plan setting.

Molecular Characterization and Landscape of Breast cancer Models from a multi-omics Perspective

Breast cancer is well-known to be a highly heterogenous disease. This facet of cancer makes finding a research model that mirrors the disparate intrinsic features challenging. With advances in multi-omics technologies, establishing parallels between the various models and human tumors is increasingly intricate. Here we review the various model systems and their relation to primary breast tumors using available omics data platforms. Among the research models reviewed here, breast cancer cell lines have the least resemblance to human tumors since they have accumulated many mutations and copy number alterations during their long use. Moreover, individual proteomic and metabolomic profiles do not overlap with the molecular landscape of breast cancer. Interestingly, omics analysis revealed that the initial subtype classification of some breast cancer cell lines was inappropriate. In cell lines the major subtypes are all well represented and share some features with primary tumors. In contrast, patient-derived xenografts (PDX) and patient-derived organoids (PDO) are superior in mirroring human breast cancers at many levels, making them suitable models for drug screening and molecular analysis. While patient derived organoids are spread across luminal, basal- and normal-like subtypes, the PDX samples were initially largely basal but other subtypes have been increasingly described. Murine models offer heterogenous tumor landscapes, inter and intra-model heterogeneity, and give rise to tumors of different phenotypes and histology. Murine models have a reduced mutational burden compared to human breast cancer but share some transcriptomic resemblance, and representation of many breast cancer subtypes can be found among the variety subtypes. To date, while mammospheres and three- dimensional cultures lack comprehensive omics data, these are excellent models for the study of stem cells, cell fate decision and differentiation, and have also been used for drug screening. Therefore, this review explores the molecular landscapes and characterization of breast cancer research models by comparing recent published multi-omics data and analysis.

Artificial intelligence: opportunities and challenges in the clinical applications of triple-negative breast cancer

Triple-negative breast cancer (TNBC) accounts for 15-20% of all invasive breast cancer subtypes. Owing to its clinical characteristics, such as the lack of effective therapeutic targets, high invasiveness, and high recurrence rate, TNBC is difficult to treat and has a poor prognosis. Currently, with the accumulation of large amounts of medical data and the development of computing technology, artificial intelligence (AI), particularly machine learning, has been applied to various aspects of TNBC research, including early screening, diagnosis, identification of molecular subtypes, personalised treatment, and prediction of prognosis and treatment response. In this review, we discussed the general principles of artificial intelligence, summarised its main applications in the diagnosis and treatment of TNBC, and provided new ideas and theoretical basis for the clinical diagnosis and treatment of TNBC.

Nasopharyngeal Carcinoma Subtype Discovery via Immune Cell Scores from Tumor Microenvironment

BACKGROUND

Nasopharyngeal carcinoma (NPC) is one of the most prevalent cancers with a poor prognosis. Immunotherapy, especially immune checkpoint blockade (ICB), is becoming a potential therapeutic choice for NPC patients. Thus, the identification of patients who could benefit from immunotherapy is clinically significant.

METHODS

The NPC expression profiles from GSE102349 were used to calculate the cell scores of the tumor microenvironment (TME). The consensus clustering method was utilized to identify the potential molecular subtypes among NPC samples. The hub genes were selected from subtype-specific genes by bioinformatics analysis. Machine learning models, including random forest (RF) and support vector machine (SVM) algorithms, were constructed to predict the immune subtype.

RESULTS

In the present study, we identified two TME subtypes among NPC patients. Patients with the S1 subtype have higher levels of immune cells, immune checkpoint genes, and prognosis. Using expression data profiles of NPC patients, we constructed machine learning models for predicting TME subtypes of NPC patients. This model consists of 8 genes (LCK, CD247, FYN, ZAP70, SH2D1A, CD3D, CD3E, and CD3G). Among them, LCK, FYN, SH2D1A, and CD3D were associated with better prognoses. Among the two constructed models, SVM exhibited a higher area under curve (AUC) of 0.977, when compared with RF (AUC = 0.966). The web server based on the constructed machine learning models will contribute to the identification of NPC patients likely to benefit from ICB therapies.

CONCLUSIONS

This study identified NPC subtypes and provided an accurate model to select individuals who are most likely to respond to ICB.

Prognostic signatures of sphingolipids: Understanding the immune landscape and predictive role in immunotherapy response and outcomes of hepatocellular carcinoma

Background

Hepatocellular carcinoma (HCC) is a complex disease with a poor outlook for patients in advanced stages. Immune cells play an important role in the progression of HCC. The metabolism of sphingolipids functions in both tumor growth and immune infiltration. However, little research has focused on using sphingolipid factors to predict HCC prognosis. This study aimed to identify the key sphingolipids genes (SPGs) in HCC and develop a reliable prognostic model based on these genes.

Methods

The TCGA, GEO, and ICGC datasets were grouped using SPGs obtained from the InnateDB portal. A prognostic gene signature was created by applying LASSO-Cox analysis and evaluating it with Cox regression. The validity of the signature was verified using ICGC and GEO datasets. The tumor microenvironment (TME) was examined using ESTIMATE and CIBERSORT, and potential therapeutic targets were identified through machine learning. Single-cell sequencing was used to examine the distribution of signature genes in cells within the TME. Cell viability and migration were tested to confirm the role of the key SPGs.

Results

We identified 28 SPGs that have an impact on survival. Using clinicopathological features and 6 genes, we developed a nomogram for HCC. The high- and low-risk groups were found to have distinct immune characteristics and response to drugs. Unlike CD8 T cells, M0 and M2 macrophages were found to be highly infiltrated in the TME of the high-risk subgroup. High levels of SPGs were found to be a good indicator of response to immunotherapy. In cell function experiments, SMPD2 and CSTA were found to enhance survival and migration of Huh7 cells, while silencing these genes increased the sensitivity of Huh7 cells to lapatinib.

Conclusion

The study presents a six-gene signature and a nomogram that can aid clinicians in choosing personalized treatments for HCC patients. Furthermore, it uncovers the connection between sphingolipid-related genes and the immune microenvironment, offering a novel approach for immunotherapy. By focusing on crucial sphingolipid genes like SMPD2 and CSTA, the efficacy of anti-tumor therapy can be increased in HCC cells.

Chromatin Regulator-Related Gene Signature for Predicting Prognosis and Immunotherapy Efficacy in Breast Cancer

Background

Many studies have found that chromatin regulators (CRs) are correlated with tumorigenesis and disease prognosis. Here, we attempted to build a new CR-related gene model to predict breast cancer (BC) survival status.

Methods

First, the CR-related differentially expressed genes (DEGs) were screened in normal and tumor breast tissues, and the potential mechanism of CR-related DEGs was determined by function analysis. Based on the prognostic DEGs, the Cox regression model was applied to build a signature for BC. Then, survival and receiver operating characteristic (ROC) curves were performed to validate the signature's efficacy and identify its independent prognostic value. The CIBERSORT and tumor immune dysfunction and exclusion (TIDE) algorithms were used to assess the immune cells infiltration and immunotherapy efficacy for this signature, respectively. Additionally, a novel nomogram was also built for clinical decisions.

Results

We identified 98 CR-related DEGs in breast tissues and constructed a novel 6 CR-related gene signature (ARID5A, ASCL1, IKZF3, KDM4B, PRDM11, and TFF1) to predict the outcome of BC patients. The prognostic value of this CR-related gene signature was validated with outstanding predictive performance. The TIDE analysis revealed that the high-risk group patients had a better response to immune checkpoint blockade (ICB) therapy.

Conclusion

A new CR-related gene signature was built, and this signature could provide the independent predictive capability of prognosis and immunotherapy efficacy for BC patients.

Machine learning links different gene patterns of viral infection to immunosuppression and immune-related biomarkers in severe burns

Introduction

Viral infection, typically disregarded, has a significant role in burns. However, there is still a lack of biomarkers and immunotherapy targets related to viral infections in burns.

Methods

Virus-related genes (VRGs) that were extracted from Gene Oncology (GO) database were included as hallmarks. Through unsupervised consensus clustering, we divided patients into two VRGs molecular patterns (VRGMPs). Weighted gene co-expression network analysis (WGCNA) was performed to study the relationship between burns and VRGs. Random forest (RF), least absolute shrinkage and selection operator (LASSO) regression, and logistic regression were used to select key genes, which were utilized to construct prognostic signatures by multivariate logistic regression. The risk score of the nomogram defined high- and low-risk groups. We compared immune cells, immune checkpoint-related genes, and prognosis between the two groups. Finally, we used network analysis and molecular docking to predict drugs targeting CD69 and SATB1. Expression of CD69 and SATB1 was validated by qPCR and microarray with the blood sample from the burn patient.

Results

We established two VRGMPs, which differed in monocytes, neutrophils, dendritic cells, and T cells. In WGCNA, genes were divided into 14 modules, and the black module was correlated with VRGMPs. A total of 65 genes were selected by WGCNA, STRING, and differential expression analysis. The results of GO enrichment analysis were enriched in Th1 and Th2 cell differentiation, B cell receptor signaling pathway, alpha-beta T cell activation, and alpha-beta T cell differentiation. Then the 2-gene signature was constructed by RF, LASSO, and LOGISTIC regression. The signature was an independent prognostic factor and performed well in ROC, calibration, and decision curves. Further, the expression of immune cells and checkpoint genes differed between high- and low-risk groups. CD69 and SATB1 were differentially expressed in burns.

Discussion

This is the first VRG-based signature (including 2 key genes validated by qPCR) for predicting survival, and it could provide vital guidance to achieve optimized immunotherapy for immunosuppression in burns.

CAPN8 involves with exhausted, inflamed, and desert immune microenvironment to influence the metastasis of thyroid cancer

BACKGROUND

Thyroid cancer (THCA) is the most prevalent malignant disease of the endocrine system, in which 5-year survival can attain about 95%, but patients with metastasis have a poor prognosis. Very little is known about the role of CAPN8 in the metastasis of THCA. In particular, the effect of CAPN8 on the tumor immune microenvironment (TIME) and immunotherapy response is unclear.

MATERIAL AND METHODS

Multiome datasets and multiple cohorts were acquired for analysis. Firstly, the expression and the prognostic value of CAPN8 were explored in public datasets and in vitro tumor tissues. Then, hierarchical clustering analysis was performed to identify the immune subtypes of THCA according to the expression of CAPN8 and the activities of related pathways. Subsequent analyses explored the different patterns of TIME, genetic alteration, DNA replication stress, drug sensitivity, and immunotherapy response among the three immune phenotypes. Finally, five individual cohorts of thyroid cancer were utilized to test the robustness and extrapolation of the three immune clusters.

RESULTS

CAPN8 was found to be a significant risk factor for THCA with a markedly elevated level of mRNA and protein in tumor tissues. This potential oncogene could induce the activation of epithelial-mesenchymal transition and E2F-targeted pathways. Three subtypes were identified for THCA, including immune exhausted, inflamed, and immune desert phenotypes. The exhausted type was characterized by a markedly increased expression of inhibitory receptors and infiltration of immune cells but was much more likely to respond to immunotherapy. The immune desert type was resistant to common chemotherapeutics with extensive genomic mutation and copy number variance.

CONCLUSION

The present study firstly explored the role of CAPN8 in the metastasis of THCA from the aspects of TIME. Three immune subtypes were identified with quite different patterns of prognosis, immunotherapy response, and drug sensitivity, providing novel insights for the treatment of THCA and helping understand the cross-talk between CAPN8 and tumor immune microenvironment.

Classification of Subgroups with Immune Characteristics Based on DNA Methylation in Luminal Breast Cancer

Luminal breast cancer (BC) accounts for a large proportion of patients in BC, with high heterogeneity. Determining the precise subtype and optimal selection of treatment options for luminal BC is a challenge. In this study, we proposed an MSBR framework that integrate DNA methylation profiles and transcriptomes to identify immune subgroups of luminal BC. MSBR was implemented both on a key module scoring algorithm and “Boruta” feature selection method by DNA methylation. Luminal A was divided into two subgroups and luminal B was divided into three subgroups using the MSBR. Furthermore, these subgroups were defined as different immune subgroups in luminal A and B respectively. The subgroups showed significant differences in DNA methylation levels, immune microenvironment (immune cell infiltration, immune checkpoint PD1/PD-L1 expression, immune cell cracking activity (CYT)) and pathology features (texture, eccentricity, intensity and tumor-infiltrating lymphocytes (TILs)). The results also showed that there is a subgroup in both luminal A and B that has the benefit from immunotherapy. This study proposed a classification of luminal BC from the perspective of epigenetics and immune characteristics, which provided individualized treatment decisions.

Construction of an Epithelial-Mesenchymal Transition-Related Model for Clear Cell Renal Cell Carcinoma Prognosis Prediction

Background. A rising amount of data demonstrates that the epithelial-mesenchymal transition (EMT) in clear cell renal cell carcinomas (ccRCC) is connected with the advancement of the cancer. In order to understand the role of EMT in ccRCC, it is critical to integrate molecules involved in EMT into prognosis prediction. The objective of this project was to establish a prognosis prediction model using genes associated with EMT in ccRCC. Methods. We acquired the mRNA expression profiles and clinical information about ccRCC from TCGA database. In this study, we measured differentially expressed EMT-related genes (DEEGs) by two comparison groups (tumor versus normal tissues; “stages I-II” versus “stages III-IV” tumor tissues). Based on classification and regression random forest models, we identified the most important DEEGs in predicting prognosis. Afterwards, a risk-score model was created using the identified important DEEGs. The prediction ability of the risk-score model was calculated by the area under the curve (AUC). A nomogram for prognosis prediction was built using the risk-score in combination with clinical factors. Results. Among the 72 DEEGs, the classification and regression random forest models identified six hub genes (DKK1, DLX4, IL6, KCNN4, RPL22L1, and SPDEF), which exhibited the highest importance values in both models. Through the expression of these six hub genes, a novel risk-score was developed for the prognosis prediction of ccRCC. ROC curves showed the risk-score performed well in both the training (0.749) and testing (0.777) datasets. According to the survival analysis, individuals who were separated into high/low-risk groups had statistically different outcomes in terms of prognosis. Besides, the risk-score model also showed outstanding ability in assessing the progression of ccRCC after treatment. In terms of nomogram, the concordance index (C-index) was 0.79. Additionally, we predicted the differences in response to chemotherapy drugs among patients from low- and high-risk groups. Conclusion. Gene signatures related to EMT could be useful in predicting ccRCC prognosis.

Developing a 5-Gene Signature Related to Pyroptosis for Osteosarcoma Patients

Although the incidence of osteosarcoma (OS) is relatively low compared with other cancer types, the overall survival of metastatic OS was less than 30%. This study aimed to reveal the role of pyroptosis in osteosarcoma and develop a prognostic model related to pyroptosis. Weighted correlation network analysis (WGCNA) was applied to identify key gene modules related to pyroptosis. Univariate Cox regression analysis was used to screen prognostic genes related to pyroptosis. The least absolute shrinkage and selection operator (LASSO) and stepwise Akaike information criterion (stepAIC) were employed to optimize and construct a prognostic model. Five prognostic genes (COL13A1, TNFRSF1A, LILRA6, CTNNBIP1, and CD180) related to pyroptosis were identified. According to the 5-gene signature, OS samples were divided into high- and low-PPRS groups with differential prognosis. Immune-related pathways were more activated in the low-PPRS group. The 5-gene signature was effective and robust to predict OS prognosis. These five prognostic genes were involved in OS development and may serve as new targets for developing therapeutic drugs.

Comprehensive Analysis of Histone Modifications in Hepatocellular Carcinoma Reveals Different Subtypes and Key Prognostic Models

Histone modification, an important epigenetic mechanism, is related to the carcinogenesis of hepatocellular carcinoma (HCC). In three datasets, we screened 88 epigenetic-dysregulated PCGs (epi-PCGs) , which were significantly associated with HCC survival and could cluster HCC into three molecular subtypes. These subtypes were associated with prognosis, immunomodulatory alterations, and response to different treatment strategies. Based on 88 epi-PCGs in the TCGA training set, a risk prediction model composed of 4 epi-PCGs was established. The model was closely related to the clinicopathological features and showed a strong predictive ability in different clinical subgroups. In addition, the risk prediction model was an independent prognostic factor for patients with HCC. The significance of epi-PCGs in HCC is revealed by our data analysis.

Identification of a Two-lncRNA Signature with Prognostic and Diagnostic Value for Hepatocellular Carcinoma

Background

Accumulating evidence has revealed the important role of long noncoding RNAs (lncRNA) in tumorigenesis and progression of hepatocellular carcinoma (HCC). This study aimed to identify potential lncRNAs that can serve as diagnostic and prognostic signatures for HCC.

Methods

Expression profiling analysis was performed to identify differentially expressed lncRNAs (DElncRNA) between HCC and matched normal samples by integrating two independent microarray datasets. Functional Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were explored by Gene Set Variation Analysis. The prognostic and diagnostic models were developed based on two DElncRNAs. Real-time PCR was used to quantify the relative expressions of candidate lncRNAs.

Results

Two robust DElncRNAs were identified and verified by quantitative PCR between HCC and matched normal samples. Function enrichment analysis revealed that they were associated with the wound healing process. The two lncRNAs were subsequently used to construct a prognostic risk model for HCC. Patients with high-risk scores estimated by the model showed a shorter survival time than low-risk patients (P < 0.001). Besides, the two lncRNA-based HCC diagnostic models exhibited good performance in discriminating HCC from normal samples on both training and test sets. The values of area under the curve (AUC) for early (I–II) and late (III–IV) HCC detection were 0.88 and 0.93, respectively.

Conclusions

The two wound healing-related DElncRNAs showed robust performance for HCC prognostic prediction and detection, implying their potential role as diagnostic and prognostic markers for HCC.

Bioinformatics Analysis Reveals an Association between Autophagy, Prognosis, Tumor Microenvironment, and Immunotherapy in Osteosarcoma

Autophagy is a catabolic pathway involved in the regulation of bone homeostasis. We explore clinical correlation of autophagy-related key molecules to establish risk signature for predicting the prognosis, tumor microenvironment (TME), and immunotherapy response of osteosarcoma. Single cell RNA sequencing data from GSE162454 dataset distinguished malignant cells from normal cells in osteosarcoma. Autophagy-related genes (ARGs) were extracted from the established risk signature of the Molecular Signatures Database of Gene Set Enrichment Analysis (GSEA) by univariate Cox and least absolute shrinkage and selection operator (LASSO) Cox regression analysis. Overall survival (OS), TME score, abundance of infiltrating immune cells, and response to immune-checkpoint blockade (ICB) treatment in patients with different risks were compared based on risk score. Nine ARGs were identified and risk signature was constructed. In all osteosarcoma datasets, the OS was significantly longer in the high-risk patients than low-risk onset. Risk signature significantly stratified clinical outcomes, including OS, metastatic status, and survival status. Risk signature was an independent variable for predicting osteosarcoma OS and showed high accuracy. A nomogram based on risk signature and metastases was developed. The calibration curve confirmed the consistency in 1-year, 3-year, and 5-year predicted OS and the actual OS. The risk score was related to 6 kinds of T cells and macrophages, myeloid-derived suppressor cell, natural killer cell, immune score, and stromal score in TME. The risk signature helped in predicting patients' response to anti-PD1/anti-PD-L1 treatment. The ARGs-led risk signature has important value for survival prediction, risk stratification, tumor microenvironment, and immune response evaluation of osteosarcoma.

Construction of ceRNA Networks Associated With CD8 T Cells in Breast Cancer

Background

The infiltration of CD8 T cells is usually linked to a favorable prognosis and may predict the therapeutic response of breast cancer patients to immunotherapy. The purpose of this research is to investigate the competing endogenous RNA (ceRNA) network correlated with the infiltration of CD8 T cells.

Methods

Based on expression profiles, CD8 T cell abundances for each breast cancer (BC) patient were inferred using the bioinformatic method by immune markers and expression profiles. We were able to extract the differentially expressed RNAs (DEmRNAs, DEmiRNAs, and DElncRNAs) between low and high CD8 T-cell samples. The ceRNA network was constructed using Cytoscape. Machine learning models were built by lncRNAs to predict CD8 T-cell abundances. The lncRNAs were used to develop a prognostic model that could predict the survival rates of BC patients. The expression of selected lncRNA (XIST) was validated by quantitative real-time PCR (qRT-PCR).

Results

A total of 1,599 DElncRNAs, 89 DEmiRNAs, and 1,794 DEmRNAs between high and low CD8 T-cell groups were obtained. Two ceRNA networks that have positive or negative correlations with CD8 T cells were built. Among the two ceRNA networks, nine lncRNAs (MIR29B2CHG, NEAT1, MALAT1, LINC00943, LINC01146, AC092718.4, AC005332.4, NORAD, and XIST) were selected for model construction. Among six prevalent machine learning models, artificial neural networks performed best, with an area under the curve (AUC) of 0.855. Patients from the high-risk category with BC had a lower survival rate compared to those from the low-risk group. The qRT-PCR results revealed significantly reduced XIST expression in normal breast samples, which was consistent with our integrated analysis.

Conclusion

These results potentially provide insights into the ceRNA networks linked with T-cell infiltration and provide accurate models for T-cell prediction.

Exploring the Interplay between Metabolism and Tumor Microenvironment Based on Four Major Metabolism Pathways in Colon Adenocarcinoma

Tumor metabolism plays a critical role in tumor progression. However, the interaction between metabolism and tumor microenvironment (TME) has not been comprehensively revealed in colon adenocarcinoma (COAD). We used unsupervised consensus clustering to establish three molecular subtypes (clusters) based on the enrichment score of four major metabolism pathways in TCGA-COAD dataset. GSE17536 was used as a validation dataset. Single-cell RNA sequencing data (GSE161277) was employed to further verify the reliability of subtyping and characterize the correlation between metabolism and TME. Three clusters were identified and they performed distinct prognosis and molecular features. Clust3 had the worst overall survival and the highest enrichment score of glycolysis. 86 differentially expressed genes (DEGs) were identified, in which 11 DEGs were associated with favorable prognosis and 75 DEGs were associated with poor prognosis. Striking correlations were observed between hypoxia and glycolysis, clust3 and hypoxia, and clust3 and angiogenesis (P < 0.001).We constructed a molecular subtyping system which was effective and reliable for predicting COAD prognosis. The 86 identified key DEGs may be greatly involved in COAD progression, and they provide new perspectives and directions for further understanding the mechanism of metabolism in promoting aggressive phenotype by interacting with TME.

Classification of Muscle Invasive Bladder Cancer to Predict Prognosis of Patients Treated with Immunotherapy

Background

Recently, immunotherapies have been approved for advanced muscle invasive bladder cancer (MIBC) treatment, but only a small fraction of MIBC patients could achieve a durable drug response. Our study is aimed at identifying tumor microenvironment (TME) subtypes that have different immunotherapy response rates.

Methods

The mRNA expression profiles of MIBC samples from seven discovery datasets (GSE13507, GSE31684, GSE32548, GSE32894, GSE48075, GSE48276, and GSE69795) were analyzed to identify TME subtypes. The identified TME subtypes were then validated by an independent dataset (TCGA-MIBC). The subtype-related biomarkers were discovered using computational analyses and then utilized to establish a random forest predictive model. The associations of TME subtypes with immunotherapy therapeutic responses were investigated in a group of patients who had been treated with immunotherapy. A prognostic index model was constructed using the subtype-related biomarkers. Two nomograms were built by the subtype-related biomarkers or the clinical parameters.

Results

Two TME subtypes, including ECM-enriched class (EC) and immune-enriched class (IC), were found. EC was associated with greater extracellular matrix (ECM) pathways, and IC was correlated with immune pathways, respectively. Overall survival was significantly greater for tumors classified as IC, whereas the EC subtype had a worse prognosis. A total of nine genes (AKAP12, APOL3, CXCL13, CXCL9, GBP4, LRIG1, PEG3, PODN, and PTPRD) were selected by computational analyses to construct the random forest model. The area under the curve (AUC) values for this model were 0.827 and 0.767 in the testing and external validation datasets, respectively. Therapeutic response rates were greater in IC patients than in EC patients (28 percent vs. 18 percent). Patients with a high prognostic index had a poorer prognosis than those with a low prognostic index. The nomogram constructed from nine genes and stage achieved a C-index of 0.71.

Conclusion

The present investigation defined two distinct TME subtypes and developed models to assess immunotherapeutic treatment outcomes.

Machine learning-based in-hospital mortality prediction of HIV/AIDS patients with Talaromyces marneffei infection in Guangxi, China

Objective

Talaromycosis is a serious regional disease endemic in Southeast Asia. In China, Talaromyces marneffei (T. marneffei) infections is mainly concentrated in the southern region, especially in Guangxi, and cause considerable in-hospital mortality in HIV-infected individuals. Currently, the factors that influence in-hospital death of HIV/AIDS patients with T. marneffei infection are not completely clear. Existing machine learning techniques can be used to develop a predictive model to identify relevant prognostic factors to predict death and appears to be essential to reducing in-hospital mortality.

Methods

We prospectively enrolled HIV/AIDS patients with talaromycosis in the Fourth People’s Hospital of Nanning, Guangxi, from January 2012 to June 2019. Clinical features were selected and used to train four different machine learning models (logistic regression, XGBoost, KNN, and SVM) to predict the treatment outcome of hospitalized patients, and 30% internal validation was used to evaluate the performance of models. Machine learning model performance was assessed according to a range of learning metrics, including area under the receiver operating characteristic curve (AUC). The SHapley Additive exPlanations (SHAP) tool was used to explain the model.

Results

A total of 1927 HIV/AIDS patients with T. marneffei infection were included. The average in-hospital mortality rate was 13.3% (256/1927) from 2012 to 2019. The most common complications/coinfections were pneumonia (68.9%), followed by oral candida (47.5%), and tuberculosis (40.6%). Deceased patients showed higher CD4/CD8 ratios, aspartate aminotransferase (AST) levels, creatinine levels, urea levels, uric acid (UA) levels, lactate dehydrogenase (LDH) levels, total bilirubin levels, creatine kinase levels, white blood-cell counts (WBC) counts, neutrophil counts, procaicltonin levels and C-reactive protein (CRP) levels and lower CD3⁺ T-cell count, CD8⁺ T-cell count, and lymphocyte counts, platelet (PLT), high-density lipoprotein cholesterol (HDL), hemoglobin (Hb) levels than those of surviving patients. The predictive XGBoost model exhibited 0.71 sensitivity, 0.99 specificity, and 0.97 AUC in the training dataset, and our outcome prediction model provided robust discrimination in the testing dataset, showing an AUC of 0.90 with 0.69 sensitivity and 0.96 specificity. The other three models were ruled out due to poor performance. Septic shock and respiratory failure were the most important predictive features, followed by uric acid, urea, platelets, and the AST/ALT ratios.

Conclusion

The XGBoost machine learning model is a good predictor in the hospitalization outcome of HIV/AIDS patients with T. marneffei infection. The model may have potential application in mortality prediction and high-risk factor identification in the talaromycosis population.

Talaromyces marneffei can cause a fatal deeply disseminated fungal infection- talaromycosis. It is widely distributed in Southeast Asia and spreading globally, the disease is insidious and responsible for significant deaths. Clinicians need easy-to-use tools to make decisions on which patients are at a higher risk of dying after infecting T. marneffei. In this study, conducted in Southern China, we have evolved XGBoost machine learning model. 15 clinical indicators and laboratory measures were used to estimate a patient’s risk of dying in the hospital due to the T. marneffei infection. The study showed that the machine learning model has good predictive ability when tested in an internal testing population of patients. We expect that the model could help clinicians assess a patient’s risk of death in just the time of admission to help decide on early treatment timing of high-risk patients who are likely to die.

Identification of Cancer-Associated Fibroblast Subtype of Triple-Negative Breast Cancer

Background

There is limited knowledge about the role of cancer-associated fibroblasts (CAF) in the tumor microenvironment of triple-negative breast cancer (TNBC).

Methods

Three hundred and thirty-five TNBC samples from four datasets were retrieved and analyzed. In order to determine the CAF subtype by combining gene expression profiles, an unsupervised clustering analysis was adopted. The prognosis, enriched pathways, immune cells, immune scores, and tumor purity were compared between CAF subtypes. The genes with the highest importance were selected by bioinformatics analysis. The machine learning model was built to predict the TNBC CAF subtype by these selected genes.

Results

TNBC samples were classified into two CAF subtypes (CAF+ and CAF-). The CAF- subtype of TNBC was linked to the longer overall survival and more immune cells than the CAF+ subtype. CAF- and CAF+ were enriched in immune-related pathways and extracellular matrix pathways, respectively. Bioinformatics analysis identified 9 CAF subtype-related markers (ADAMTS12, AEBP1, COL10A1, COL11A1, CXCL11, CXCR6, EDNRA, EPPK1, and WNT7B). We constructed a robust random forest model using these 9 genes, and the area under the curve (AUC) value of the model was 0.921.

Conclusion

The current study identified CAF subtypes based on gene expression profiles and found that CAF subtypes have significantly different overall survival, immune cells, and immunotherapy response rates.