Discovery and Validation of Key Biomarkers based on Machine

Learning and Immune Infiltrates in Ovarian Cancer

Linlin Zhang

1,*

, Mingming Yu

1,*

, Xuehua Bi

, Guanglei Yu

and Kai Zhao

School of Software, Xinjiang University, Urumqi 830091, China

College of Biomedical Engineering and Technology, Xinjiang Medical University, Urumqi 830011, China

College of Information Science and Engineering, Xinjiang University, Urumqi 830046, China

Keywords:

Ovarian Cancer, Immune Cell, Feature Selection, Machine Learning, Biomarkers, CIBERSOFT.

Abstract:

Ovarian cancer (OC) is the deadliest gynecological malignancy which survival rate mainly depends on early

detection. Our purpose was to search for potential OC diagnostic markers and to examine the role of immune

cell infiltration in its disease process. OC expression profiles were extracted from Gene Expression Omnibus

(GEO) and differentially expressed genes (DEGs) were identified with the limma R package and subjected to

functional correlation analysis. We used Hilbert-Schmidt Independence Criterion Least Absolute Shrinkage

and Selection Operator (HSIC-Lasso), Support Vector Machine-Recursive Feature Elimination (SVM-RFE)

algorithms and Minimum Redundancy Maximum Relevance (mRMR) to select gene features and chose the

random forest (RF) algorithm as the classifier to validate the results of gene selection. Finally, we utilized

CIBERSORT to bulk gene ex-pression profiles of OC for quantifying 22 subsets of immune cells.

Subsequently, we analysed the correlation between diagnostic markers and infiltrating immune cells. ABCA8,

IGFBP2 and REEP1 were identified as diagnostic markers for OC in this study (AUC=0.96), and a total of

380 DEGs were identified. Immune cell infiltration analysis showed that plasma cells, CD8 T cells and

activated memory CD4 T cells may be involved in the occurrence and development of OC. In addition,

ABCA8 was positively correlated with neutrophils, monocytes, activated NK cells while negatively correlated

with activated CD4 memory T cells, naïve B cells and macrophages M1. IGFBP2 was positively correlated

with macrophages M1 while negatively correlated with monocytes and neutrophils. REEP1 was positively

correlated with neutrophils, monocytes, macrophages M2, activated NK cells and plasma cells while

negatively correlated with resting NK cells, activated CD4 memory T cells and CD8 T cells. In conclusion,

ABCA8, IGFBP2 and REEP1 can be used as diagnostic markers of OC, and immune cell infiltration plays a

crucial role in the occurrence and progression of OC.

1 INTRODUCTION

Ovarian cancer is one of the most common

malignancies in women and the leading cause of

death from gynecologic cancers, ranking fifth in the

United States. In 2022, the United States is estimated

to have about 19,880 new cases and 12,810 deaths

(SIEGEL, 2022). Owing to the particularity of

ovarian location, most cases of ovarian cancer are

diagnosed as advanced and have metastasized in the

abdomen. Ovarian cancer has poor prognosis and

high mortality due to lack of effective methods for

early detection (JAYSON, 2014). Biomarkers can be

utilized to develop personalized therapeutic

interventions, and the treatment of tumors is

increasingly being influenced by biomarkers. Finding

effective biomarkers and studying their roles in the

occurrence and development of OC are of great

significance for elucidating the pathogenesis,

diagnosis, prognosis of ovarian cancer. The

development of microarray-based analysis and high-

throughput biological sequencing technology have

made it possible to analyze DEGs in order that

biomarkers related to cancer diagnosis, treatment and

prognosis can be identified and potential biological

mechanisms discovered (VOGELSTEIN, 2013).

GEO contains high-throughput gene expression data

submitted by institutions that can be uploaded or

downloaded by other researchers (CLOUGH, 2016).

Based on the high-throughput gene expression data

provided by GEO, we can deeply insight into the

biological functions and regulatory mechanisms of

Zhang, L., Yu, M., Bi, X., Yu, G. and Zhao, K.

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer.

DOI: 10.5220/0012019200003633

In Proceedings of the 4th International Conference on Biotechnology and Biomedicine (ICBB 2022), pages 253-265

ISBN: 978-989-758-637-8

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

253

ovarian cancer, clarify the mechanisms of its

occurrence and development, and explore new

diagnostic and therapeutic approaches. The GEO

database provides data support for our research.

In this study, 7 microarray datasets (GSE10971,

GSE18520, GSE26712, GSE36668, GSE40595,

GSE54388 and GSE69428) were extracted from the

GEO. Firstly, the DEGs between OC and normal

ovary samples were identified based on the above

seven datasets, and their potential biological

functions were analyzed by functional and path-way

enrichment analysis. Then, we validated OC

diagnostic markers using machine learning

algorithms and analyzed immune cell infiltration into

OC tissue using CIBERSOFT. Finally, we performed

correlation analysis to explore the relationship

between three diagnostic markers and immune cell

infiltration.

2 MATERIALS AND METHODS

2.1 Datasets Selection and Data

Processing

Seven microarray datasets (GSE10971, GSE18520,

GSE26712, GSE36668, GSE40595, GSE54388 and

GSE69428) were extracted from the GEO database

(https://www.ncbi.nlm.nih.gov/gds/). GSE26712

was based on Affymetrix GPL96 plat-form

(Affymetrix human genome U133A array),

GSE10971, GSE18520, GSE36668, GSE40595,

GSE54388 and GSE69428 were based on Affymetrix

GPL570 platform (Affymetrix human genome U133

Plus 2.0 array). A total of 426 tissue samples, in-

cluding 348 OC samples and 78 normal ovarian

samples.

In this study, six microarray datasets (GSE10971,

GSE26712, GSE36668, GSE40595, GSE54388 and

GSE69428) were used the limma R package

(RITCHIE, 2015) for filtering batch effects due to

datasets combination. Before removing batch effects,

we used log2 to transform the expressed values of the

dataset.

2.2 Functional Correlation Analysis

In this study, we utilized clusterProfiler R package

(YU, 2012) for Gene Ontology (GO), Disease

Oncology (DO) and KEGG enrichment analysis. GO

annotation are grouped into three categories:

biological process (BP), cellular component (CC) and

molecular function (MF). The enriched KEGG

pathway and GO annotations with P < 0.05 were

selected.

For gene set enrichment analysis (GSEA), we

obtained the GSEA software (version 3.0) from the

GSEA website, divided the samples into two groups

based on the expression levels of 3 hub genes, and

downloaded the background gene set required for the

study from the Molecular Signatures Database v7.4.

Based on gene expression profile and phenotype

grouping, we set the minimum gene to 5, the

maximum gene to 5000, and one thousand resamples,

P value < 0.05 and FDR < 0.25 were considered

statistically significant.

2.3 Gene Selection

In this study, we utilized singular value

decomposition (SVD) to process the expression

matrix. Subsequently, SVM-RFE is used to filter the

optimal feature subset, it gets the importance of each

feature by the importance of the feature, eliminates

the least important features from the current feature

set, and repeats this process recursively on the set

after elimination until finally reaching the number of

features to be selected (SUYKENS, 1999).

Algorithm 1: SVM-RFE

Input: Original gene sets F = (f

, f

, …, f

), number

of targets: k

Output: Target genes 𝐹

∗

=(𝑓



∗

,𝑓



∗

,…,𝑓



∗

)

1 Initialize F

= F;

2 Train SVM according to F

and get the descending

ranking of all features;

3 Delete the last feature and update F

;

4 If the number of F

is equal to k, ends; otherwise,

return to step 2.

The mRMR selects features based on mutual

information with the aim of finding the set of genes

in the original set of genes that are most correlated

with the final output result but least correlated with

each other (HANCHUAN P, 2005). In gene set S, the

maximally important and minimally redundant gene

i* is given by:

𝑖

∗

=𝑎𝑟𝑔𝑚𝑎𝑋



∈

𝑅



𝑄

,

(

)

Where 𝑅



represents the maximum correlation

condition, 𝑄

,

represents the minimum redundancy

condition.

HSIC Lasso is a kernel-based feature selection

algorithm. It focuses on the nonlinear correlation

between input features and output results, and finally

finds non redundant features that are highly

dependent on output results (YAMADA, 2014).

ICBB 2022 - International Conference on Biotechnology and Biomedicine

254

min

∈ℝ



𝐿 −𝛼



𝐾

(



)











+𝜆

‖

𝛼

‖



(

)

s.t.𝛼



,···,𝛼



≥0,

Where

‖

∙

‖



is the Frobenius norm, K

()

H , L=HLH, H∈ℝ

×

are centered

matrices, λ is the regularization parameter, α∈ℝ



is a parameter to be sought with non-negative

constraints.

2.4 Classification and Evaluation

Metrics

For evaluating the result of feature selection in the

previous section, we chose the random forest

algorithm as the classifier.

This study used accuracy, precision, recall,

specificity, and AUC values as the criteria for

determining the results of the experiment. The

definitions are as follows:

𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦=

𝑇𝑃+ 𝑇𝑁

𝑇𝑃+ 𝐹𝑃+ 𝑇𝑁+ 𝐹𝑁

(

)

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛=

𝑇𝑃

𝑇𝑃+ 𝐹𝑃

(

)

𝑅𝑒𝑐𝑎𝑙𝑙=

𝑇𝑃

𝑇𝑃+ 𝐹𝑁

(

)

𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦=

𝑇𝑁

𝐹𝑃+ 𝑇𝑁

(

)

AUC is defined as the area composed of receiver

operating characteristic (ROC) curve and abscissa,

which can intuitively evaluate the performance of the

classifier. The higher AUC is, the better the

classification ability of the model is.

2.5 Analysis of Immune Cell

Infiltration

CIBERSOFT is a gene-based deconvolution

algorithm that evaluates the relative proportion of 22

tumor infiltrating immune cell profiles based on

expression files, covering plasma cells, B cells, T

cells, and myeloid cell subsets. In this research, we

utilized CIBERSORT to obtain an immune cell

infiltration matrix, filtering out samples with p<0.05.

Then, we used PCA to analyse the matrix data and

draw a PCA cluster diagram. To visualize the

correlation of the 22 infiltrating immune cells, we

produced a correlation heatmap using the corrplot R

package (FRIENDLY, 2002). Violin plots were

drawn using the ggplot2 R package to display the

differences in immune cell infiltration.

3 RESULTS

3.1 Data Processing and Differential

Expression Analysis

It is necessary to remove the batch effect from the

gene expression matrix and normalize after 6 datasets

(GSE10971, GSE26712, GSE36668, GSE40595,

GSE54388 and GSE69428) were merged. The

detailed flowchart was presented in Figure 1. We

obtained 380 DEGs using the limma R package after

data processing, as shown in the volcano plot (Figure

2).

Table 1: Details of GEO datasets.

Dataset Samples Features OC Normal

GSE10971 37 23520 13 24

GSE18520 63 23519 53 10

GSE26712 195 13515 185 10

GSE36668 12 23520 8 4

GSE40595 77 23520 63 14

GSE54388 22 23520 16 6

GSE69428 20 18184 10 10

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer

255

Figure 1: Flowchart of the integrated analysis.

Figure 2: Volcano plot of DEGs; red represents up-regulated differential genes, black represents no significant difference

genes, and green represents down-regulated differential genes.

3.2 Functional Correlation Analysis

The DO results show that the DEGs were mainly

involved in embryonal cancer, ovarian cancer,

embryoma, germ cell cancer, female reproductive

organ cancer (Figure 3). In the KEGG pathway

enrichment analysis, the up-regulated DEGs mainly

participated in cell cycle, DNA replication, oocyte

meiosis, p53 signaling pathway (Figure 4A). The GO

analysis of DEGs classified DEGs into three

functional groups: BP, MF, and CC. For the BP

group, the up-regulated DEGs were mainly involved

in chromosome segregation, mitotic nuclear division,

nuclear division (Figure 4B). For the MF group, the

up-regulated DEGs mainly participated in cyclin-

dependent protein serine/threonine kinase regulator

activity, microtubule binding, tubulin binding

(Figure 4C). For the CC group, the up-regulated

DEGs mainly participated in spindle, chromosomal

region, mitotic spindle (Figure 4D). GSEA results

mainly included TGF-β signaling, hedgehog

signaling and epithelial-mesenchymal transition

(Figure 5).

ICBB 2022 - International Conference on Biotechnology and Biomedicine

256

Figure 3: Histogram of DO analysis. The horizontal axis represents the number of DEGs under the DO item.

Figure 4: Histogram of KEGG and GO analysis. The horizontal axis represents the number of DEGs under the GO and KEGG

item.

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer

257

Figure 5: Gene set enrichment analysis. P value < 0.05 and FDR < 0.25 were considered statistically significant.

TGF_BETA_SIGNALING (FDR = 0.0983), HEDGEHOG_SIGNALING (FDR = 0.1169) and

EPITHELIAL_MESENCHYMAL_TRANSITION (FDR = 0.1669).

3.3 Screening and Identification of Key

Biomarkers

We downloaded seven datasets from GEO. The

number of patients was 363 (295 in the OC group; 68

in the control group). To search the biomarkers of

OC, we performed gene selection through SVM-

RFE, HSIC Lasso and mRMR (DENG, 2020;

MARVI-KHORASANI, 2019). The evaluation

metrics of the three feature selection algorithms were

shown in Table 2. The results of the SVM-RFE

showed that 11 genes were identified as signature

genes in OC. Meanwhile, we utilized the HSIC Lasso

algorithm to screen out 10 characteristic genes in OC.

Finally, the mRMR feature selection algorithm

selected 18 key genes related to ovarian cancer

(Figure 6). Known from the Venn diagram of the

three algorithms, three diagnosis-related genes were

obtained (Figure 6). We used GSE18520 to verify the

diagnostic efficacy of ABCA8, IGFBP2 and REEP1,

and the AUC results showed that the combination of

the three genes can reach a very high level in the

verification set (AUC = 0.96). The results showed

that ABCA8, IGFBP2 and REEP1 had greater

diagnostic value (Figure 7).

Table 2: Evaluation metrics for different feature selection algorithms.

Metrics SVM-RFE HSIC Lasso mRMR

Features 11 10 18

Accuracy (%) 96.00 98.13 97.87

Precision (%) 99.39 99.86 99.88

Recall (%) 97.40 98.31 98.08

Specificity (%) 89.60 96.49 96.56

AUC (%) 98.30 99.50 99.58

ICBB 2022 - International Conference on Biotechnology and Biomedicine

258

Figure 6: Venn diagram and average accuracy of three feature selection algorithms.

Figure 7: The ROC curve of patient category verification after the combination of three diagnostic markers.

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer

259

3.4 Immune Cell Infiltration

To explore the role of immune cell infiltration in OC,

in this study, we utilized PCA cluster analysis to

observe differences between OC samples and normal

samples, and we found significant differences in

immune cell infiltration between the two groups.

(Figure 8A). Derived from the correlation heat map

of immune cells, activated CD4 memory T cells,

resting mast cells, and activated dendritic cells had a

significant positive correlation. The violin plot

indicated that the fraction for plasma cells, T cells

CD8 and T cells CD4 memory activated in the OC

group were significantly higher than the normal

group. On the contrary, the fractions of many cells

were lower than the normal group, such as T cells

CD4 naïve, T cells gamma delta, monocytes, and

neutrophils (Figure 8B).

The correlation analysis between ABCA8,

IGFBP2, REEP1 and infiltrating immune cells

showed that ABCA8 was positively correlated with

neutrophils, monocytes and activated NK cells while

negatively correlated with activated CD4 memory T

cells, naïve B cells and macrophages M1. IGFBP2

was positively correlated with macrophages M1

while negatively correlated with monocytes and

neutrophils. REEP1 was positively correlated with

neutrophils, monocytes, macrophages M2, activated

NK cells and plasma cells while negatively correlated

with resting NK cells, activated CD4 memory T cells

and CD8 T cells (Figure 9).

Figure 8: Immune cell infiltration analysis. (A) PCA cluster plot of immune cell infiltration between OC samples and normal

samples. (B) Violin diagram of the proportion of 22 types of immune cells. (C) Correlation heat map of 22 types of immune

cells.

ICBB 2022 - International Conference on Biotechnology and Biomedicine

260

Figure 9: Correlation between ABCA8, IGFBP2, REEP1 and infiltrating immune cell.

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer

261

4 DISCUSSION

Ovarian cancer is the deadliest gynecological

malignancies. In 2018, there were 295,414 new cases

and 184,799 deaths worldwide, showing a significant

upward trend. Due to the onset of ovarian cancer is

very insidious, there are no obvious symptoms in the

early stage of the disease, and accurate methods for

early screening are lacking. As a result, more than

70% of patients are at an advanced stage at initial

diagnosis. Studies have shown that the infiltration of

immune cells plays a crucial role in the occurrence

and development of OC (ZHANG, 2020). Therefore,

it is of great significance to use machine learning to

find specific markers and analyze the infiltration

patterns of OC immune cells to improve the

prognosis of OC patients. The CIBERSOFT tool also

facilitates the analysis of disease facial cell

infiltration. In this research, we utilized machine

learning method to identify the diagnostic markers of

OC. In addition, we also analyzed the role of immune

cell infiltration in OC.

First, we assembled 7 OC gene expression

datasets from GEO database, with a total of 426

samples, including 348 OC samples and 78 normal

samples. We identified 380 DEGs using limma R

package. KEGG results show that DEGs mainly

participated in the KEGG pathway including cell

cycle, DNA replication, oocyte meiosis, p53

signaling pathway. GO enrichment results show that

DEGs were mainly related to chromosome

segregation, mitotic nuclear division and cyclin-

dependent protein serine/threonine kinase regulator

activity. DO enrichment results show that the

diseases mainly include embryonal cancer, ovarian

cancer, embryoma, germ cell cancer, female

reproductive organ cancer. Furthermore, GSEA

results mainly involves TGF-β signaling, hedgehog

signaling and epithelial-mesenchymal transition.

Research by Basu et al. (BASU, 2015) found that the

activation of TGF-β signaling can induce the invasion

of OC cells. Wen et al. (WEN, 2020) showed that the

self-renewal, migration and invasion of OC stem

cells can be inhibited by blocking TGF-β signaling.

The study of Doheny et al. (DOHENY, 2020)

concluded that abnormal activation of the hedgehog

signaling plays a crucial role in the occurrence and

development of ovarian cancer. Nieto et al. (NIETO,

2016) showed that epithelial-mesenchymal transition

is the main process for the transformation of early

ovarian tumors into aggressive and metastatic

malignancies. The above results show that the

analytical results of our study are precise.

SVD is a widely used algorithm in machine

learning, mainly applied for feature decomposition in

dimension reduction algorithms. SVM-RFE is one of

the commonly used feature selection methods. The

so-called recursive feature removal is to take the form

of a loop to get the order of features. In each recursive

process, the score of each feature will be calculated

according to certain rules, remove the lowest score

(the least important feature), then repeat the process,

until all genes have their own sequence. HSIC is a

non-linear feature selection method that considers the

relationship between non-linear input and output.

HSIC Lasso uses HSIC to measure the dependencies

between variables. The mRMR selects features based

on mutual information with the aim of finding the set

of genes in the original set of genes that are most

correlated with the final output result but least

correlated with each other. First, we utilize singular

value decomposition to process the expression

matrix. The genes are then filtered using three

algorithms to create the optimal classification model.

Finally, combining the screening results of the three

algorithms, ABCA8, IGFBP2 and REEP1 were

identified as diagnostic markers for OC.

The ABC transporter superfamily can mediate the

ATP-dependent transport of many exogenous and

endogenous substances through the lipid bilayer. The

ABC transporter is responsible for the transport of

various inflammatory mediators and lipids. These

substances are directly related to tumor progression

in ovarian cancer. Therefore, they can contribute to

the clinical outcome and become a potential

therapeutic target for OC. Hedditch et al.

(HEDDITCH, 2014) showed that ABCA transporter

correlated with poor prognosis in serous ovarian

cancer, suggesting that lipid trafficking was a

potentially important process in epithelial ovarian

cancer. Cancer cells rely on de novo synthesis of

lipids to produce fatty acids to meet the increased

energy requirements of tumor growth. More and

more evidences indicate that lipid metabolism is

dysregulated in cancers including ovarian cancer

(PYRAGIUS, 2013). Therefore, we believe that

ABCA8 may be involved in the pathological process

of OC. The IGFBP family plays a vital role in

regulating basic biological activities outside and

inside cells (BAXTER, 2014). Research by Lee et al.

(LEE, 2005) found through western blotting and

tissue microarray analysis that IGFBP2 was

significantly overexpressed in malignant ovarian

tissues, indicating that IGFBP2 enhanced the

invasive ability of ovarian cancer cells. In addition,

the increase in IGFBP2 expression is positively

correlated with the level of serum tumor marker

ICBB 2022 - International Conference on Biotechnology and Biomedicine

262

CA125 (FLYVBJERG, 1997). Therefore, we

conclude that IGFBP2 can be used as a potential

marker for the diagnosis of OC, and given that

IGFBP2 can enhance the invasion ability of cancer

cells, IGFBP2 has great potential as a therapeutic

target in the future. REEP1 is a member of the

endoplasmic reticulum (ER)-forming protein family

that localizes to the ER and the plasma membrane

(RENVOISé, 2016; BJöRK, 2013). Voloshanenko et

al. (VOLOSHANENKO, 2018) used independent

experiments to prove that REEP1 can be used as a

non-classical target gene in colon cancer cells. Zhao

et al. (ZHAO, 2019) showed that REEP1 can be used

as a molecular diagnostic marker and therapeutic

target for breast cancer. GO annotations related to

REEP1 include microtubule binding, and research

showed that Paclitaxel, as a microtubule inhibitor,

can be used to treat high-grade serous ovarian cancer.

Due to the tumor resistance of paclitaxel as a

therapeutic drug, the discovery of new microtubule

inhibitors has become more and more urgent.

Therefore, it is necessary to study the mechanism of

the occurrence and development of REEP1 in ovarian

cancer.

We used CIBERSOFT to further explore the role

of immune cell infiltration in OC. The results showed

that an increased infiltration of plasma cells, CD8 T

cells, activated memory CD4 T cells, and a decreased

infiltration of naïve CD4 T cells, gamma delta T cells,

monocytes, neutrophils. Kroeger et al. (KROEGER ,

2016) indicated that plasma cells are related with

CD8(+) tumor-infiltrating lymphocytes response in

OC. Sato et al. (SATO, 2005) used

immunohistochemical analysis to confirm that

intraepithelial CD8+ tumor infiltrating lymphocytes

and high CD8+/regulatory T cell ratio can

significantly improve the survival rate of ovarian

cancer patients. In the tumor-associated lymphocytes

of ascites in patients with ovarian cancer, the

frequency of activated memory T-cells was

significantly increased, and they may be activated by

the tumor or the tumor-related microenvironment

(LANDSKRON, 2015). The patient with high mast

cell infiltration had a longer survival period was

found in (CHAN, 2005). The relationship between

macrophages and ovarian cancer was explored by

Zhang et al. (ZHANG, 2020), and they had found that

macrophages promote the proliferation and migration

of ovarian tumors, providing a potential treatment

method for patients with ovarian cancer. A new

mechanism of action for paclitaxel was discovered by

Wanderley et al. (WANDERLEY, 2018) through the

formation of several tumor models, they showed that

paclitaxel switched tumor associated macrophages to

an M1 like antitumor phenotype by reactivating

anticancer immune responses, which provided a

rationale for a new treatment regimen combining

paclitaxel with immunotherapy. Regarding

monocytes, Prat et al. (PRAT, 2020) verified that

they can be used as a biomarker of ascites immune

status and ovarian cancer progression. The

prerequisite step before the metastasis of ovarian

cancer in situ is that neutrophils flow into the

omentum, and the extracellular traps of neutrophils

combine with ovarian cancer cells to promote

metastasis (LEE, 2019). The above analysis shows

that plasma cells, CD8 T cells, activated memory

CD4 T cells, monocytes, neutrophils play crucial

roles in the pathogenesis of OC and should be the

focus of further research.

We combined SVD with three feature selection

algorithms to validate diagnostic markers for OC and

analysed immune cell infiltration in OC tissues using

CIBERSOFT. Our study has certain limitations due

to the limited genetic data available for analysis.

5 CONCLUSIONS

In this paper, we found that ABCA8, IGFBP2 and

REEP1 are diagnostic markers of OC. Also, this

study found that plasma cells, CD8 T cells, and

activated memory CD4 T cells may be involved in

the occurrence and development of OC. Besides,

ABCA8 was positively correlated with neutrophils,

monocytes, activated NK cells and negatively

correlated with activated CD4 memory T cells, naïve

B cells, macrophages M1. IGFBP2 was positively

correlated with macrophages M1 and negatively

correlated with monocytes, neutrophils. REEP1 was

positively correlated with neutrophils, monocytes,

macrophages M2, activated NK cells, plasma cells

and negatively correlated with resting NK cells,

activated CD4 memory T cells, CD8 T cells. In the

future, the role of these immune cells in ovarian

cancer requires further research to identify the targets

of OC immunotherapy, which can help to improve

the level of immunomodulatory therapy for OC

patients.

ACKNOWLEDGMENTS

This work was supported in part by the Natural

Science Foundation of Xinjiang Uygur Autonomous

Region under Grant 2019D01C062, 2019D01C041,

2019D01C205, and 2020D01C028; in part by the

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer

263

National Natural Science Foundation of China under

Grant 12061071; in part by the Higher Education of

Xinjiang Uygur Autonomous Region under Grant

XJEDU2020Y003, and XJEDU2019Y006; in part by

the Major Science and Technology Special Project of

Xinjiang Uygur Autonomous Region under Grant

2020A02001-1; in part by the Major Science and

Technology Project of Sichuan Science and

Technology Plan under Grant 2020YFQ0018; in part

by the National Innovation Training Project for

College Student under Grant 202010755021.

REFERENCES

Basu M, Bhattacharya R, Ray U, et al. Invasion of ovarian

cancer cells is induced byPITX2-mediated activation of

TGF-β and Activin-A[J]. Molecular cancer, 2015,

14(162.

Baxter R C. IGF binding proteins in cancer: mechanistic

and clinical insights[J]. Nature reviews Cancer, 2014,

14(5): 329-341.

Björk S, Hurt C M, Ho V K, et al. REEPs are membrane

shaping adapter proteins that modulate specific g

protein-coupled receptor trafficking by affecting ER

cargo capacity[J]. PloS one, 2013, 8(10): e76366.

Chan J K, Magistris A, Loizzi V, et al. Mast cell density,

angiogenesis, blood clotting, and prognosis in women

with advanced ovarian cancer[J]. Gynecologic

oncology, 2005, 99(1): 20-25.

Clough E, Barrett T. The Gene Expression Omnibus

Database[J]. Methods in molecular biology (Clifton,

NJ), 2016, 1418: 93-110.

Deng Y-J, Ren E-H, Yuan W-H, et al. GRB10 and E2F3 as

Diagnostic Markers of Osteoarthritis and Their

Correlation with Immune Infiltration[J]. Diagnostics,

2020, 10(3): 171-187.

Doheny D, Manore S G, Wong G L, et al. Hedgehog

Signaling and Truncated GLI1 in Cancer[J]. Cells,

2020, 9(9): 2114-2130.

Flyvbjerg A, Mogensen O, Mogensen B, et al. Elevated

serum insulin-like growth factor-binding protein 2

(IGFBP-2) and decreased IGFBP-3 in epithelial

ovarian cancer: correlation with cancer antigen 125 and

tumor-associated trypsin inhibitor[J]. The Journal of

clinical endocrinology and metabolism, 1997, 82(7):

2308-2313.

Friendly M. Corrgrams: Exploratory Displays for

Correlation Matrices[J]. The American Statistician,

2002, 56(4): 316-324.

Guangchuang Yu L-G W, Yanyan Han, and Qing-Yu He.

clusterProfiler: an R Package for Comparing Biological

Themes Among Gene Clusters[J]. OMICS: A Journal

of Integrative Biology, 2012, 16(5): 284-287.

Hanchuan P, Fuhui L, Ding C. Feature selection based on

mutual information criteria of max-dependency, max-

relevance, and min-redundancy[J]. IEEE Transactions

on Pattern Analysis and Machine Intelligence, 2005,

27(8): 1226-1238.

Hedditch E L, Gao B, Russell A J, et al. ABCA Transporter

Gene Expression and Poor Outcome in Epithelial

Ovarian Cancer[J]. JNCI: Journal of the National

Cancer Institute, 2014, 106(7): dju149.

Jayson G C, Kohn E C, Kitchener H C, et al. Ovarian

cancer[J]. The Lancet, 2014, 384(9951): 1376-1388.

KROEGER D R, MILNE K, NELSON B H. Tumor-

Infiltrating Plasma Cells Are Associated with Tertiary

Lymphoid Structures, Cytolytic T-Cell Responses, and

Superior Prognosis in Ovarian Cancer[J]. Clinical

cancer research: an official journal of the American

Association for Cancer Research, 2016, 22(12): 3005-

3015.

Landskron J, Helland Ø, Torgersen K M, et al. Activated

regulatory and memory T-cells accumulate in

malignant ascites from ovarian carcinoma patients [J].

Cancer Immunology, Immunotherapy, 2015, 64(3):

337-347.

Lee W, Ko S Y, Mohamed M S, et al. Neutrophils facilitate

ovarian cancer premetastatic niche formation in the

omentum [J]. The Journal of experimental medicine,

2019, 216(1): 176-194.

Lee E J, Mircean C, Shmulevich I, et al. Insulin-like growth

factor binding protein 2 promotes ovarian cancer cell

invasion[J]. Molecular cancer, 2005, 4(1): 7-13.

Marvi-Khorasani H, Usefi H. Feature Clustering Towards

Gene Selection[C]. In Proc. of the 18th IEEE

International Conference on Machine Learning and

Applications (ICMLA). Dec 16-19, 2019, Boca Raton,

FL, USA, pp: 1466-1469.

Nieto M A, Huang R Y, Jackson R A, et al. EMT: 2016[J].

Cell, 2016, 166(1): 21-45.

Prat M, Le Naour A, Coulson K, et al. Circulating

CD14(high) CD16(low) intermediate blood monocytes

as a biomarker of ascites immune status and ovarian

cancer progression[J]. Journal for immunotherapy of

cancer, 2020, 8(1): e000472.

Pyragius C E, Fuller M, Ricciardelli C, et al. Aberrant lipid

metabolism: an emerging diagnostic and therapeutic

target in ovarian cancer[J]. International journal of

molecular sciences, 2013, 14(4): 7742-7756.

Renvoisé B, Malone B, Falgairolle M, et al. Reep1 null

mice reveal a converging role for hereditary spastic

paraplegia proteins in lipid droplet regulation[J].

Human molecular genetics, 2016, 25(23): 5111-5125.

Ritchie M E, Phipson B, Wu D, et al. limma powers

differential expression analyses for RNA-sequencing

and microarray studies[J]. Nucleic Acids Research,

2015, 43(7): e47.

Sato E, Olson S H, Ahn J, et al. Intraepithelial CD8+ tumor-

infiltrating lymphocytes and a high CD8+/regulatory T

cell ratio is associated with favorable prognosis in

ovarian cancer[J]. Proceedings of the National

Academy of Sciences of the United States of America,

2005, 102(51): 18538-18543.

Siegel R L, Miller K D, Fuchs H E, et al. Cancer statistics,

2022[J]. CA: A Cancer Journal for Clinicians, 2022,

72(1): 7-33.

ICBB 2022 - International Conference on Biotechnology and Biomedicine

264

Suykens J A K, Vandewalle J. Least Squares Support

Vector Machine Classifiers[J]. Neural Processing

Letters, 1999, 9(3): 293-300.

Vogelstein B, Papadopoulos N, Velculescu V E, et al.

Cancer Genome Landscapes[J]. Science, 2013,

339(6127): 1546-1358.

Voloshanenko O, Schwartz U, Kranz D, et al. β-catenin-

independent regulation of Wnt target genes by RoR2

and ATF2/ATF4 in colon cancer cells[J]. Scientific

Reports, 2018, 8(1): 3178.

Wanderley C W, Colón D F, Luiz J P M, et al. Paclitaxel

Reduces Tumor Growth by Reprogramming Tumor-

Associated Macrophages to an M1 Profile in a TLR4-

Dependent Manner[J]. Cancer research, 2018, 78(20):

5891-5900.

Wen H, Qian M, He J, et al. Inhibiting of self-renewal,

migration, and invasion of ovarian cancer stem cells by

blocking TGF-β pathway[J]. PloS one, 2020, 15(3):

e0230230.

Yamada M, Jitkrittum W, Sigal L, et al. High-Dimensional

Feature Selection by Feature-Wise Kernelized

Lasso[J]. Neural Computation, 2014, 26(1): 185-207.

Zhang Q F, Li J, Jiang K, et al. CDK4/6 inhibition promotes

immune infiltration in ovarian cancer and synergizes

with PD-1 blockade in a B cell-dependent manner[J].

Theranostics, 2020, 10(23): 10619-10633.

Zhao C, Lou Y, Wang Y, et al. A gene expression

signature-based nomogram model in prediction of

breast cancer bone metastases[J]. Cancer Medicine,

2019, 8(1): 200-208.

Zhang Q, Li H, Mao Y, et al. Apoptotic SKOV3 cells

stimulate M0 macrophages to differentiate into M2

macrophages and promote the proliferation and

migration of ovarian cancer cells by activating the ERK

signaling pathway[J]. International journal of

molecular medicine, 2020, 45(1): 10-22.

Discovery and Validation of Key Biomarkers based on Machine Learning and Immune Inﬁltrates in Ovarian Cancer

265