Gang Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Gang Liu has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Lei Liu
Department of Medical and Pharmaceutical Informatics, Hebei Medical University
10 shared publications Aoshen Wu
Institutes of Biomedical Sciences and Intelligent Medicine Institute, Fudan University, Shanghai 200032, China
4 shared publicationsXueqing Peng
Institute of Biomedical Sciences, Fudan University, Shanghai, P.R. China
4 shared publicationsYunhe Liu
Institutes of Biomedical Sciences Fudan University Shanghai China
4 shared publicationsHe Ren
Faculty of Medical Instrumentation, Shanghai University of Medicine and Health Sciences, Shanghai 201318, China
3 shared publications- Yunhe Liu
Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center, Houston, USA
3 shared publications
Latest Publications
- Open Access
- 31 July 2025
- Published by Springer Nature in BMC Medical Genomics
- Vol. 18 (1) , 1-16
A lipid metabolism-related gene signature predicts prognosis after tamoxifen treatment in ER + breast cancer and reflects tumor microenvironment heterogeneity through single-cell analysis
Although the clinical outcome of ER + breast cancer patients receiving tamoxifen after surgery is favorable, a proportion of patients experience recurrence or death due to disease progression. In this study, by integrating lipid metabolism gene expression and machine learning data, a prognostic model based on gene expression was developed using the TCGA-ER + BRCA dataset (N = 183) and validated with the GSE17705 (N = 298), GSE22219 (N = 134), GSE42568 (N = 70), and GSE58644 (N = 147) datasets. Patients were stratified into high- and low-risk groups based on the median risk score of the signature. Comparative analyses of survival, genomic features, immune infiltration, and drug sensitivity were performed between these groups. Patients in the high-risk group had worse survival outcomes than those in the low-risk group. The five-year overall survival AUC of the model was 0.858, indicating good performance. High-risk patients were characterized by USH2A and KMT2C mutations, genomic amplification, and enriched JAK-STAT pathway and cytokine-cytokine receptor interaction pathways. Resting CD4 + memory T cells, activated mast cells, and myeloid dendritic cells were significantly enriched in the low-risk group, while M0 macrophages were enriched in the high-risk group. Single-cell sequencing analyses also revealed that the model was significantly associated with macrophages and the percentage of proliferating myeloid cells. The signature was also associated with sensitivity to multiple drugs. Cell–cell interaction difference analyses suggested that cancer-related signaling pathways, especially the SIRPα/CD47/IL6 pathway, were decreased in high-risk patients, but these samples exhibited increased SPP1 interactions. The signature captures lipid metabolic reprogramming and immunosuppression, providing a biomarker for prognosis and precision therapy in tamoxifen-treated ER + breast cancer.
- 31 July 2025
- Published by Springer Nature in BMC Medical Genomics
- Vol. 18 (1) , 1-16
Open AccessA lipid metabolism-related gene signature predicts prognosis after tamoxifen treatment in ER + breast cancer and reflects tumor microenvironment heterogeneity through single-cell analysis
Although the clinical outcome of ER + breast cancer patients receiving tamoxifen after surgery is favorable, a proportion of patients experience recurrence or death due to disease progression. In this study, by integrating lipid metabolism gene expression and machine learning data, a prognostic model based on gene expression was developed using the TCGA-ER + BRCA dataset (N = 183) and validated with the GSE17705 (N = 298), GSE22219 (N = 134), GSE42568 (N = 70), and GSE58644 (N = 147) datasets. Patients were stratified into high- and low-risk groups based on the median risk score of the signature. Comparative analyses of survival, genomic features, immune infiltration, and drug sensitivity were performed between these groups. Patients in the high-risk group had worse survival outcomes than those in the low-risk group. The five-year overall survival AUC of the model was 0.858, indicating good performance. High-risk patients were characterized by USH2A and KMT2C mutations, genomic amplification, and enriched JAK-STAT pathway and cytokine-cytokine receptor interaction pathways. Resting CD4 + memory T cells, activated mast cells, and myeloid dendritic cells were significantly enriched in the low-risk group, while M0 macrophages were enriched in the high-risk group. Single-cell sequencing analyses also revealed that the model was significantly associated with macrophages and the percentage of proliferating myeloid cells. The signature was also associated with sensitivity to multiple drugs. Cell–cell interaction difference analyses suggested that cancer-related signaling pathways, especially the SIRPα/CD47/IL6 pathway, were decreased in high-risk patients, but these samples exhibited increased SPP1 interactions. The signature captures lipid metabolic reprogramming and immunosuppression, providing a biomarker for prognosis and precision therapy in tamoxifen-treated ER + breast cancer.
- Open Access
- 25 July 2024
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 25 (5)
A comparison of scRNA-seq annotation methods based on experimentally labeled immune cell subtype dataset
Cell-type annotation is a critical step in single-cell data analysis. With the development of numerous cell annotation methods, it is necessary to evaluate these methods to help researchers use them effectively. Reference datasets are essential for evaluation, but currently, the cell labels of reference datasets mainly come from computational methods, which may have computational biases and may not reflect the actual cell-type outcomes. This study first constructed an experimentally labeled immune cell-subtype single-cell dataset of the same batch and systematically evaluated 18 cell annotation methods. We assessed those methods under five scenarios, including intra-dataset validation, immune cell-subtype validation, unsupervised clustering, inter-dataset annotation, and unknown cell-type prediction. Accuracy and ARI were evaluation metrics. The results showed that SVM, scBERT, and scDeepSort were the best-performing supervised methods. Seurat was the best-performing unsupervised clustering method, but it couldn't fully fit the actual cell-type distribution. Our results indicated that experimentally labeled immune cell-subtype datasets revealed the deficiencies of unsupervised clustering methods and provided new dataset support for supervised methods.
- 25 July 2024
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 25 (5)
Open AccessA comparison of scRNA-seq annotation methods based on experimentally labeled immune cell subtype dataset
Cell-type annotation is a critical step in single-cell data analysis. With the development of numerous cell annotation methods, it is necessary to evaluate these methods to help researchers use them effectively. Reference datasets are essential for evaluation, but currently, the cell labels of reference datasets mainly come from computational methods, which may have computational biases and may not reflect the actual cell-type outcomes. This study first constructed an experimentally labeled immune cell-subtype single-cell dataset of the same batch and systematically evaluated 18 cell annotation methods. We assessed those methods under five scenarios, including intra-dataset validation, immune cell-subtype validation, unsupervised clustering, inter-dataset annotation, and unknown cell-type prediction. Accuracy and ARI were evaluation metrics. The results showed that SVM, scBERT, and scDeepSort were the best-performing supervised methods. Seurat was the best-performing unsupervised clustering method, but it couldn't fully fit the actual cell-type distribution. Our results indicated that experimentally labeled immune cell-subtype datasets revealed the deficiencies of unsupervised clustering methods and provided new dataset support for supervised methods.
- Open Access
- 15 July 2024
- Published by Baishideng Publishing Group Inc. in World Journal of Gastrointestinal Oncology
- Vol. 16 (7) , 3032-3054
Integrated single-cell and bulk RNA sequencing revealed an epigenetic signature predicts prognosis and tumor microenvironment colorectal cancer heterogeneity
BACKGROUND Colorectal cancer (CRC) prognosis prediction is currently a major challenge. Epigenetic regulation has been widely reported for its role in cancer development. AIM To construct a robust prognostic signature, we used developed and validated across datasets. METHODS After constructing the signature, the prognostic value of the signature was evaluated in the TCGA cohort and six independent datasets (GSE17526, GSE17537, GSE33113, GSE37892, GSE39048 and GSE39582). The clinical, genomic and transcriptomic features related to the signature were identified. The correlations of the signature score with immune cell infiltration and cell-cell interactions were analyzed. The correlations between the signature score and the sensitivity to different drugs were also predicted. RESULTS In the TCGA cohort, patients in the low-risk group according to the signature score had longer survival than those in the high-risk group, and this finding was validated in the validation datasets. The signature was a prognostic factor independent of age and sex and was correlated with stage and PD-1 /PD-L1 expression. Area under the receiving operating characteristic curve was 0.72. Genomic association analyses revealed that samples from high-risk patients exhibited chromosomal instability. Transcriptomic analyses revealed that the signature score was significantly associated with multiple cellular pathways. Bulk RNA-seq and single-cell sequencing data revealed that the signature reflected differences in infiltrating immune cell-tumor cell interactions, especially for macrophages. The signature also predicted the putative drug sensitivity of CRC samples. CONCLUSION The signature is a valuable biomarker for predicting CRC prognosis and reflects multiple features of CRC, especially macrophage infiltration in the microenvironment.
- 15 July 2024
- Published by Baishideng Publishing Group Inc. in World Journal of Gastrointestinal Oncology
- Vol. 16 (7) , 3032-3054
Open AccessIntegrated single-cell and bulk RNA sequencing revealed an epigenetic signature predicts prognosis and tumor microenvironment colorectal cancer heterogeneity
BACKGROUND Colorectal cancer (CRC) prognosis prediction is currently a major challenge. Epigenetic regulation has been widely reported for its role in cancer development. AIM To construct a robust prognostic signature, we used developed and validated across datasets. METHODS After constructing the signature, the prognostic value of the signature was evaluated in the TCGA cohort and six independent datasets (GSE17526, GSE17537, GSE33113, GSE37892, GSE39048 and GSE39582). The clinical, genomic and transcriptomic features related to the signature were identified. The correlations of the signature score with immune cell infiltration and cell-cell interactions were analyzed. The correlations between the signature score and the sensitivity to different drugs were also predicted. RESULTS In the TCGA cohort, patients in the low-risk group according to the signature score had longer survival than those in the high-risk group, and this finding was validated in the validation datasets. The signature was a prognostic factor independent of age and sex and was correlated with stage and PD-1 /PD-L1 expression. Area under the receiving operating characteristic curve was 0.72. Genomic association analyses revealed that samples from high-risk patients exhibited chromosomal instability. Transcriptomic analyses revealed that the signature score was significantly associated with multiple cellular pathways. Bulk RNA-seq and single-cell sequencing data revealed that the signature reflected differences in infiltrating immune cell-tumor cell interactions, especially for macrophages. The signature also predicted the putative drug sensitivity of CRC samples. CONCLUSION The signature is a valuable biomarker for predicting CRC prognosis and reflects multiple features of CRC, especially macrophage infiltration in the microenvironment.
- Open Access
- 8 June 2023
- Published by Elsevier in Acta Histochemica
- Vol. 125 (6) , 152057
Classification of colorectal cancer consensus molecular subtypes using attention-based multi-instance learning network on whole-slide images
- 8 June 2023
- Published by Elsevier in Acta Histochemica
- Vol. 125 (6) , 152057
Open AccessClassification of colorectal cancer consensus molecular subtypes using attention-based multi-instance learning network on whole-slide images
- Open Access
Attention-Based Deep Multiple-Instance Learning for Classifying Circular RNA and Other Long Non-Coding RNA
Circular RNA (circRNA) is a distinguishable circular formed long non-coding RNA (lncRNA), which has specific roles in transcriptional regulation, multiple biological processes. The identification of circRNA from other lncRNA is necessary for relevant research. In this study, we designed attention-based multi-instance learning (MIL) network architecture fed with a raw sequence, to learn the sparse features of RNA sequences and to accomplish the circRNAs identification task. The model outperformed the state-of-art models. Moreover, following the validation of the attention mechanism effectiveness by the handwritten digit dataset, the key sequence loci underlying circRNA’s recognition were obtained based on the corresponding attention score. Then, motif enrichment analysis identified some of the key motifs for circRNA formation. In conclusion, we designed deep learning network architecture suitable for learning gene sequences with sparse features and implemented it for the circRNA identification task, and the model has strong representation capability in the indication of some key loci.
Open AccessAttention-Based Deep Multiple-Instance Learning for Classifying Circular RNA and Other Long Non-Coding RNA
Circular RNA (circRNA) is a distinguishable circular formed long non-coding RNA (lncRNA), which has specific roles in transcriptional regulation, multiple biological processes. The identification of circRNA from other lncRNA is necessary for relevant research. In this study, we designed attention-based multi-instance learning (MIL) network architecture fed with a raw sequence, to learn the sparse features of RNA sequences and to accomplish the circRNAs identification task. The model outperformed the state-of-art models. Moreover, following the validation of the attention mechanism effectiveness by the handwritten digit dataset, the key sequence loci underlying circRNA’s recognition were obtained based on the corresponding attention score. Then, motif enrichment analysis identified some of the key motifs for circRNA formation. In conclusion, we designed deep learning network architecture suitable for learning gene sequences with sparse features and implemented it for the circRNA identification task, and the model has strong representation capability in the indication of some key loci.
- Open Access
- 15 September 2021
- Published by Wiley in Journal of Cellular and Molecular Medicine
- Vol. 25 (20) , 9617-9626
Identification of metabolism‐associated molecular subtype in ovarian cancer
Ovarian cancer (OC) is the most lethal gynaecological cancer with genomic complexity and extensive heterogeneity. This study aimed to characterize the molecular features of OC based on the gene expression profile of 2752 previously characterized metabolism‐relevant genes and provide new strategies to improve the clinical status of patients with OC. Finally, three molecular subtypes (C1, C2 and C3) were identified. The C2 subtype displayed the worst prognosis, upregulated immune‐cell infiltration status and expression level of immune checkpoint genes, lower burden of copy number gains and losses and suboptimal response to targeted drug bevacizumab. The C1 subtype showed downregulated immune‐cell infiltration status and expression level of immune checkpoint genes, the lowest incidence of BRCA mutation and optimal response to targeted drug bevacizumab. The C3 subtype had an intermediate immune status, the highest incidence of BRCA mutation and a secondary optimal response to bevacizumab. Gene signatures of C1 and C2 subtypes with an opposite expression level were mainly enriched in proteolysis and immune‐related biological process. The C3 subtype was mainly enriched in the T cell‐related biological process. The prognostic and immune status of subtypes were validated in the Gene Expression Omnibus (GEO) dataset, which was predicted with a 45‐gene classifier. These findings might improve the understanding of the diversity and therapeutic strategies for OC.
- 15 September 2021
- Published by Wiley in Journal of Cellular and Molecular Medicine
- Vol. 25 (20) , 9617-9626
Open AccessIdentification of metabolism‐associated molecular subtype in ovarian cancer
Ovarian cancer (OC) is the most lethal gynaecological cancer with genomic complexity and extensive heterogeneity. This study aimed to characterize the molecular features of OC based on the gene expression profile of 2752 previously characterized metabolism‐relevant genes and provide new strategies to improve the clinical status of patients with OC. Finally, three molecular subtypes (C1, C2 and C3) were identified. The C2 subtype displayed the worst prognosis, upregulated immune‐cell infiltration status and expression level of immune checkpoint genes, lower burden of copy number gains and losses and suboptimal response to targeted drug bevacizumab. The C1 subtype showed downregulated immune‐cell infiltration status and expression level of immune checkpoint genes, the lowest incidence of BRCA mutation and optimal response to targeted drug bevacizumab. The C3 subtype had an intermediate immune status, the highest incidence of BRCA mutation and a secondary optimal response to bevacizumab. Gene signatures of C1 and C2 subtypes with an opposite expression level were mainly enriched in proteolysis and immune‐related biological process. The C3 subtype was mainly enriched in the T cell‐related biological process. The prognostic and immune status of subtypes were validated in the Gene Expression Omnibus (GEO) dataset, which was predicted with a 45‐gene classifier. These findings might improve the understanding of the diversity and therapeutic strategies for OC.
- Open Access
Attention-based deep multiple instances learning for classifying circular RNA and other long non-coding RNA
Circular RNA (circRNA) is a distinguishable circular formed long non-coding RNA (lncRNA), which has specific roles in transcriptional regulation, multiple biological processes. The identification of circRNA from other lncRNA is necessary for relevant research. In this study, we designed attention-based multi-instance learning (MIL) network architecture, which can be fed with raw sequence, to learn the sparse features in sequences and accomplish the identification task for circRNAs. The model outperformed previously reported models. Following the effectiveness validation of the attention score by the handwritten digit dataset, the key sequence loci underlying circRNAs recognition were obtained based on the corresponding attention score. Moreover, the motif enrichment analysis of the extracted key sequences identified some of the key motifs for circRNA formation. In conclusion, we designed a deep learning network architecture suitable for gene sequence learning with sparse features and implemented to the circRNA identification, and the network has a strong representation capability with its indication of some key loci.
Open AccessAttention-based deep multiple instances learning for classifying circular RNA and other long non-coding RNA
Circular RNA (circRNA) is a distinguishable circular formed long non-coding RNA (lncRNA), which has specific roles in transcriptional regulation, multiple biological processes. The identification of circRNA from other lncRNA is necessary for relevant research. In this study, we designed attention-based multi-instance learning (MIL) network architecture, which can be fed with raw sequence, to learn the sparse features in sequences and accomplish the identification task for circRNAs. The model outperformed previously reported models. Following the effectiveness validation of the attention score by the handwritten digit dataset, the key sequence loci underlying circRNAs recognition were obtained based on the corresponding attention score. Moreover, the motif enrichment analysis of the extracted key sequences identified some of the key motifs for circRNA formation. In conclusion, we designed a deep learning network architecture suitable for gene sequence learning with sparse features and implemented to the circRNA identification, and the network has a strong representation capability with its indication of some key loci.
- Open Access
Single-Cell Transcriptome Profiling Simulation Reveals the Impact of Sequencing Parameters and Algorithms on Clustering
Despite the scRNA-seq analytic algorithms developed, their performance for cell clustering cannot be quantified due to the unknown “true” clusters. Referencing the transcriptomic heterogeneity of cell clusters, a “true” mRNA number matrix of cell individuals was defined as ground truth. Based on the matrix and the actual data generation procedure, a simulation program (SSCRNA) for raw data was developed. Subsequently, the consistency between simulated data and real data was evaluated. Furthermore, the impact of sequencing depth and algorithms for analyses on cluster accuracy was quantified. As a result, the simulation result was highly consistent with that of the actual data. Among the clustering algorithms, the Gaussian normalization method was the more recommended. As for the clustering algorithms, the K-means clustering method was more stable than K-means plus Louvain clustering. In conclusion, the scRNA simulation algorithm developed restores the actual data generation process, discovers the impact of parameters on classification, compares the normalization/clustering algorithms, and provides novel insight into scRNA analyses.
Open AccessSingle-Cell Transcriptome Profiling Simulation Reveals the Impact of Sequencing Parameters and Algorithms on Clustering
Despite the scRNA-seq analytic algorithms developed, their performance for cell clustering cannot be quantified due to the unknown “true” clusters. Referencing the transcriptomic heterogeneity of cell clusters, a “true” mRNA number matrix of cell individuals was defined as ground truth. Based on the matrix and the actual data generation procedure, a simulation program (SSCRNA) for raw data was developed. Subsequently, the consistency between simulated data and real data was evaluated. Furthermore, the impact of sequencing depth and algorithms for analyses on cluster accuracy was quantified. As a result, the simulation result was highly consistent with that of the actual data. Among the clustering algorithms, the Gaussian normalization method was the more recommended. As for the clustering algorithms, the K-means clustering method was more stable than K-means plus Louvain clustering. In conclusion, the scRNA simulation algorithm developed restores the actual data generation process, discovers the impact of parameters on classification, compares the normalization/clustering algorithms, and provides novel insight into scRNA analyses.
- Open Access
Single-cell transcriptome profiling simulation reveals the impact of sequencing parameters and algorithms on clustering
Despite of scRNA-seq analytic algorithms developed, their performance for cell clustering cannot be quantified due to the unknown “true” clusters. Referencing the transcriptomic heterogeneity of cell clusters, a “true” mRNA number matrix of cell individuals was defined as ground truth. Based on the matrix and real data generation procedure, a simulation program (SSCRNA) for raw data was developed. Subsequently, the consistence between simulated data and real data was evaluated. Furthermore, the impact of sequencing depth, and algorithms for analyses on cluster accuracy was quantified. As a result, the simulation result is highly consistent with that of the real data. It is found that mis-classification rate can be attributed to multiple reasons on current scRNA platforms, and clustering accuracy is not only sensitive to sequencing depth increasement, but can also be reflected by the position of the cluster on TSNE plot. Among the clustering algorithms, Gaussian normalization method is more appropriate for current workflows. In the clustering algorithms, k-means&louvain clustering method performs better in dimension reduced data than full data, while k-means clustering method is stable under both situations. In conclusion, the scRNA simulation algorithm developed restores the real data generation process, discovered impact of parameters on mis-clustering, compared the normalization/clustering algorithms and provided novel insight into scRNA analyses.
Open AccessSingle-cell transcriptome profiling simulation reveals the impact of sequencing parameters and algorithms on clustering
Despite of scRNA-seq analytic algorithms developed, their performance for cell clustering cannot be quantified due to the unknown “true” clusters. Referencing the transcriptomic heterogeneity of cell clusters, a “true” mRNA number matrix of cell individuals was defined as ground truth. Based on the matrix and real data generation procedure, a simulation program (SSCRNA) for raw data was developed. Subsequently, the consistence between simulated data and real data was evaluated. Furthermore, the impact of sequencing depth, and algorithms for analyses on cluster accuracy was quantified. As a result, the simulation result is highly consistent with that of the real data. It is found that mis-classification rate can be attributed to multiple reasons on current scRNA platforms, and clustering accuracy is not only sensitive to sequencing depth increasement, but can also be reflected by the position of the cluster on TSNE plot. Among the clustering algorithms, Gaussian normalization method is more appropriate for current workflows. In the clustering algorithms, k-means&louvain clustering method performs better in dimension reduced data than full data, while k-means clustering method is stable under both situations. In conclusion, the scRNA simulation algorithm developed restores the real data generation process, discovered impact of parameters on mis-clustering, compared the normalization/clustering algorithms and provided novel insight into scRNA analyses.
- Open Access
Untargeted Metabolomic Characterization of Ovarian Tumors
Diagnosis of ovarian cancer is difficult due to the lack of clinical symptoms and effective screening algorithms. In this study, we aim to develop models for ovarian cancer diagnosis by detecting metabolites in urine and plasma samples. Ultra-high-performance liquid chromatography and quadrupole time-of-flight mass spectrometry (UHPLC-QTOF-MS) in positive ion mode was used for metabolome quantification in 235 urine samples and 331 plasma samples. Then, Urine and plasma metabolomic profiles were analyzed by univariate and multivariate statistics. Four groups of samples: normal control, benign, borderline and malignant ovarian tumors were enrolled in this study. A total of 1330 features and 1302 features were detected from urine and plasma samples respectively. Based on two urine putative metabolites, five plasma putative metabolites and five urine putative metabolites, three models for distinguishing normal-ovarian tumors, benign-malignant (borderline + malignant) and borderline-malignant ovarian tumors were developed respectively. The AUC (Area Under Curve) values were 0.987, 0876 and 0.943 in discovery set and 0.984, 0.896 and 0.836 in validation set for three models. Specially, the diagnostic model based on 5 plasma putative metabolites had better early-stage diagnosis performance than CA125 alone. The AUC values of the model were 0.847 and 0.988 in discovery and validation set respectively. Our results showed that normal and ovarian tumors have unique metabolic signature in urine and plasma samples, which shed light on the ovarian cancer diagnosis and classification.
Open AccessUntargeted Metabolomic Characterization of Ovarian Tumors
Diagnosis of ovarian cancer is difficult due to the lack of clinical symptoms and effective screening algorithms. In this study, we aim to develop models for ovarian cancer diagnosis by detecting metabolites in urine and plasma samples. Ultra-high-performance liquid chromatography and quadrupole time-of-flight mass spectrometry (UHPLC-QTOF-MS) in positive ion mode was used for metabolome quantification in 235 urine samples and 331 plasma samples. Then, Urine and plasma metabolomic profiles were analyzed by univariate and multivariate statistics. Four groups of samples: normal control, benign, borderline and malignant ovarian tumors were enrolled in this study. A total of 1330 features and 1302 features were detected from urine and plasma samples respectively. Based on two urine putative metabolites, five plasma putative metabolites and five urine putative metabolites, three models for distinguishing normal-ovarian tumors, benign-malignant (borderline + malignant) and borderline-malignant ovarian tumors were developed respectively. The AUC (Area Under Curve) values were 0.987, 0876 and 0.943 in discovery set and 0.984, 0.896 and 0.836 in validation set for three models. Specially, the diagnostic model based on 5 plasma putative metabolites had better early-stage diagnosis performance than CA125 alone. The AUC values of the model were 0.847 and 0.988 in discovery and validation set respectively. Our results showed that normal and ovarian tumors have unique metabolic signature in urine and plasma samples, which shed light on the ovarian cancer diagnosis and classification.