Sage Journals: Discover world-class research

Abstract

BACKGROUND:

Patients with oral squamous carcinoma (OSCC) present difficulty in precise diagnosis and poor prognosis.

OBJECTIVE:

We aimed to identify the diagnostic and prognostic indicators in OSCC and provide basis for molecular mechanism investigation of OSCC.

METHODS:

We collected sequencing data and clinical data from TCGA database and screened the differentially expressed mRNAs (DEmRNAs) and lncRNAs (DElncRNAs) in OSCC. Machine learning and modeling were performed to identify the optimal diagnostic markers. In order to determine lncRNAs with prognostic value, survival analysis was performed through combing the expression profiles with the clinical data. Finally, co-expressed DEmRNAs of lncRNAs were identified by interacted network construction and functional annotated by GO and KEGG analysis.

RESULTS:

A total of 1114 (345 up- and 769 down-regulated) DEmRNAs and 156 (86 up- and 70 down-regulated) DElncRNAs were obtained in OSCC. Following the machine learning and modeling, 15 lncRNAs were identified to be the optimal diagnostic indicators of OSCC. Among them, FOXD2.AS1 was significantly associated with survival rate of patients with OSCC. In addition, Focal adhesion and ECM-receptor interaction pathways were found to be involved in OSCC.

CONCLUSIONS

: FOXD2.AS1 might be a prognostic marker for OSCC and our study may provide more information to the further study in OSCC.

Keywords

Oral squamous carcinoma lncRNA diagnosis prognosis biomarker machine learning

1. Introduction

Oral squamous cell carcinoma (OSCC), a common type of head and neck squamous cell carcinoma (HNSCC), was the eighth most common cancer worldwide, and over 300,000 new OSCC were diagnosed each year [1]. Early detection followed by appropriate treatment could improve the cure rates and quality of life of patients with OSCC [2, 3]. Nevertheless, early oral cancer lesions were often asymptomatic and even diagnosis by general population screening was with false-positive [4, 5]. Thus, the National Institute of Dental and Craniofacial Research and The Oral Cancer Foundation have recommended that research efforts focus on developing novel detection techniques.

Next-generation high-throughput sequencing technologies have been used to generate large amounts of transcript sequences and gene expression data for patients with complicated diseases [6]. Up to date, large amounts of biomarkers were screened in salivary of patients with OSCC, including squamous cell carcinoma antigen (SCCAg) [7], carcino-embryonic antigen (CEA) [8], cyfra 21-1 [9], anti-p53 antibodies [10], interleukin-6 (IL-6) and TNF-a [11] in salivary proteomics, and mutated p53 [12], DNA polymorphism of IL-1b [13], IL-6 [14], IL-8 [15], TNF-a [16] and VEGF [17] and aberrant expressions of CD44 [18] in salivary genomics. Furthermore, a study analyzed salivary microbiota in oral cancer indicated that high salivary counts of C. gingivalis, P. melaninogenica and S. mitis might be diagnostic indicators of OSCC [19]. A study showed that lncRNA, HOTAIR, was differentially expressed in the saliva of the OSCC metastatic patients compared with that of primary cancer controls [20]. However, these biomarkers were discovered with small sample size, and several biomarkers that merely reflected changes of biological variability might not be the “true biomarkers” [21]. Thus, diagnostic indicators with higher accuracy should be identified. In addition, patients with OSCC generally had a poor prognosis, and the 5-year survival rate of OSCC remained less than 50% [22, 23, 24]. Therefore, identification of prognostic markers should also be crucial in investigation of OSCC.

Machine learning was the sub-field artificial intelligence which focused on methods to construct computer programs that learn from experience with respect to some class of tasks and a performance measure [25]. Machine learning methods have been applied to a huge variety of problems in genomics and genetics [26]. In our study, we firstly selected differentially expressed mRNAs (DEmRNAs) and lncRNAs (DElncRNAs) of OSCC in TCGA database and then screened key diagnostic and prognostic lncRNAs by machine learning. Finally, following the DElncRNA-DEmRNA interacted network construction, functional annotation was performed in the co-expressed DEmRNAs of lncRNAs. Our study identified the diagnostic and prognostic markers by combining the gene expression profile and clinical data and provided a molecular basis to the further investigation of OSCC.

2. Materials and methods

2.1 Eligible data of HNSCC screening in TCGA database

The pre-treated grade 3 RNA-Seq data and the clinical data related to HNSCC from the database set-up to October 10 ${}^{\text{th}}$ 2019 were downloaded from the TCGA database (tcga-data.nci.nih.gov/). Then, the data in oral clinical specimens (alveolar ridge, buccal mucosa, mouth bottom, tongue, lip, mouth and hard palate) were retained, and the data in specimens of pharynx, larynx, oropharynx and tonsils were excluded. Follow-up analysis was performed on patients with OSCC who had a long follow-up period.

2.2 Identification of DEmRNAs and DElncRNAs in OSCC

Clean data was obtained following exclusion of undetectable mRNAs and lncRNAs (a read count value $=$ 0 in $>$ 20% samples or in $>$ 20% normal samples). Differentially expressed analysis on clean data was then performed by R (vision 3.5.3) packages DESeq2. DEmRNAs and DElncRNAs were identified with FDR $<$ 0.01 and abs (log ${}_{2}$ FC) $>$ 2.

2.3 Identification of optimal diagnostic lncRNA biomarkers for OSCC

Modeling and machine learning algorithm were used to identify optimal diagnostic lncRNA biomarkers for OSCC. Firstly, feature selection procedures were performed: the LASSO algorithm analysis was conducted using the ‘glmnet’ package (https://cran.r-project.org/web/packages/glmnet/) to decrease data dimensions, and the optimal parameters were then selected, resulting in the smallest error rate. Secondly, the importance of each selected DElncRNA was ranked from large to small according to the value of Mean Decrease Accuracy by the random forest algorithm (random Forest; https://cran.r-project.org/web/packages/ randomForest/); the optimum number of features was indicated by adding one DElncRNA at a time in the top down method by using random forest algorithm and 10-fold cross-validation. Finally, the classification models were established by Random Forest package, Support Vector Machine (SVM, https://cran.r-project.org/web/packages/e1071/index.html) and Decision Tree (rpart, https://cran.r-project.org/web/packages /rpart/index.html), respectively. The diagnostic ability of these three models and each lncRNA biomarker was evaluated by the receiver operating characteristic (ROC) area under curve (AUC), sensitivity and specificity.

Figure 1.

Hierarchical clustering analysis of top100 DEmRNAs (A) and all DElncRNAs (B) in OSCC. Row and column represented tissue samples and DEmRNAs/DElncRNAs, respectively.

Table 1

OSCC patients characteristics ( $n=$ 331)

Parameter	Patients ( $n=$ 331)
Gender
Male	127
Female	104
Age
$\leqslant$ 50	59
$>$ 50	272
Race
Asian	9
Black or african american	22
White	289
Unknown/unavailable	11
Tumor grades
G1	51
G2	202
G3	68
G4	2
GX	6
Unavailable	2
Status
Alive	215
Dead	116

2.4 Correlation between the optimal diagnostic lncRNAs and clinical features

To further investigate the optimal diagnostic DElncRNAs for HNSCC, boxplot analysis of the correlation between tumor grades and the expression of lncRNAs were performed. In addition, survival analysis was performed to evaluate the prognostic value of optimal lncRNAs.

2.5 Co-expression analysis of the optimal diagnostic lncRNAs and mRNAs

To predict the function of the optimal diagnostic lncRNAs, correlation analysis between lncRNAs and mRNAs was analyzed using Pearson’s correlation coefficient. The co-expression pairs were obtained with $|r|>$ 0.5 and $P<$ 0.05.

2.6 Functional analysis of co-expressed mRNAs of the optimal diagnostic lncRNAs

To further predict the function of the optimal diagnostic lncRNAs, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes analysis of co-expressed DEmRNAs of which were performed using GeneCoDis3 (http://genecodis.cnb.csic.es/analysis). The threshold was set at FDR $<$ 0.05.

3. Results

3.1 Eligible data collection from TCGA database

In total, 331 cases of OSCC and 44 normal controls were collected from TCGA database, including the clinical data, expression profiles of mRNAs and lncRNAs. Characters of patients, including the OSCC grades and survival rate, were shown on Table 1.

Figure 2.

Identification of lncRNA biomarkers for OSCC. (A) Importance value of each DElncRNA ranked according to the mean decrease accuracy. (B) Variance rate of classification performance when increasing numbers of the predictive DElncRNAs.

3.2 Identification of DEmRNAs and DElncRNAs between OSCC and normal controls

After DESeq2 analysis and filtering, 1114 (345 up- and 769 down-regulated) DEmRNAs and 156 (86 up- and 70 down-regulated) DElncRNAs were obtained between patients with OSCC and normal controls. Hierarchical clustering analysis of the top 100 DEmRNAs and all DElncRNAs was presented in Fig. 1A and B, respectively.

3.3 Identification of the optimal diagnostic lncRNAs

A total of 30 diagnostic DElncRNAs were selected following the data dimensions decreasing and ranked according to Mean Decrease Accuary (Table 2, Fig. 2A). A 10-fold cross-validation result demonstrated that the accuracy reached the maximum when the number of lncRNAs was 15 (Fig. 2B). Therefore, the top15 DElncRNAs were selected for the optimal diagnostic lncRNAs. Hierarchical clustering analysis of the 15 lncRNAs was presented in Fig. 3.

Table 2
Differentially expressed lncRNAs between OSCC and normal tissues following reduced dimensions of the data

	log ${}_{2}$ (fold change)		$P$ -value	Regulation
RP11.159F24.6	3.	458798	3.59E-71	Up
RP11.863P13.3	4.	248121	2.40E-64	Up
FOXD2.AS1	2.	231206	2.42E-59	Up
CTA.384D8.35	2.	966089	2.52E-55	Up
RNF144A.AS1	3.	095125	7.36E-51	Up
ST3GAL4.AS1	2.	193601	1.31E-42	Up
RP11.760H22.2	$-$ 3.	08781	2.19E-37	Down
RP11.339B21.10	2.	237438	3.66E-30	Up
RP11.426C22.4	2.	624578	1.28E-29	Up
CTC.297N7.9	$-$ 2.	17135	9.68E-28	Down
RP11.10L7.1	$-$ 2.	35892	1.47E-26	Down
RP11.321G12.1	$-$ 3.	14645	1.74E-25	Down
AC011738.4	2.	699794	1.32E-24	Up
AP001347.6	$-$ 2.	24301	3.68E-24	Down
LA16c.380A1.1	$-$ 2.	35869	1.90E-23	Down
RP6.191P20.4	2.	183969	2.76E-23	Up
ADAMTS9.AS1	$-$ 2.	70944	2.92E-23	Down
LINC01554	$-$ 2.	86959	3.64E-22	Down
LINC01405	$-$ 3.	74222	1.39E-21	Down
SMC2.AS1	$-$ 2.	32813	7.18E-21	Down
RP11.68I18.10	$-$ 2.	93129	2.87E-20	Down
RP11.37C7.3	2.	061934	2.93E-20	Up
RP11.38M8.1	2.	238911	3.19E-19	Up
RP11.7K24.3	$-$ 2.	03269	1.06E-18	Down
MIR2117HG	2.	616516	4.63E-18	Up
AC093159.1	$-$ 2.	75372	4.10E-15	Down
DSG1.AS1	$-$ 2.	24877	1.57E-14	Down
CTC.490G23.2	$-$ 2.	59794	4.12E-14	Down
LINC00491	2.	162386	1.77E-12	Up
LINC01629	2.	087123	4.78E-07	Up

${}^{*}$ Bold highlight indicates 15 optimal diagnostic lncRNA biomarkers for OSCC.

Figure 3.

Hierarchical clustering analysis of the optimal 15 lncRNAs. Row and column represented tissue samples and lncRNAs, respectively.

Figure 4.

ROC curves of the classification models. The models were based on random forest (A), decision tree model (B) and support vector machine model (C). The value of AUC indicated the performance of the model. ROC, receiver operating characteristic; AUC, area under the curve.

Then, 15 optimal diagnostic lncRNAs were then used to establish the classification models by random Forest package, SVM and decision tree. High AUC value indicated the 15 lncRNAs were correlated to OSCC. The AUC of the random forests model was 0.996 and the specificity and sensitivity of this model were 95.5% and 99.7%, respectively (Fig. 4A). The AUC of the decision tree model was 0.835 and the specificity and sensitivity of this model were 72.7% and 96.4%, respectively (Fig. 4B). The AUC of the SVM model was 0.994, and the specificity and sensitivity of this model were 95.5% and 97.9% (Fig. 4C).

Figure 5.

Correlation between 15 biomarkers and clinical data. (A) Box-plot displayed the expression levels of (I) FOXD2.AS1, (II) RP11.159F24.6, (III) RP11.863P13.3, (IV) CTA.384D8.35, (V) RNF144A.AS1, (VI) ST3GAL4.AS1, (VII) RP11.760H22.2, (VIII) RP11.339B21.10, (IX) RP11.426C22.4, (X) CTC.297N7.9, (XI) RP11.10L7.1, (XII) LA16c.380A1.1, (XIII) ADAMTS9.AS1, (XIV) RP11.7K24.3, and (XV) CTC.490G23.2 in normal tissues and OSCC tissues of various tumor grades.

Figure 5.

Continued. (B) Survival curves of (I) FOXD2.AS1, (II) RP11.159F24.6, (III) RP11.863P13.3, (IV) CTA.384D8.35, (V) RNF144A.AS1, (VI) ST3GAL4.AS1, (VII) RP11.760H22.2, (VIII) RP11.339B21.10, (IX) RP11.426C22.4, (X) CTC.297N7.9, (XI) RP11.10L7.1, (XII) LA16c.380A1.1, (XIII) ADAMTS9.AS1, (XIV) RP11.7K24.3, and (XV) CTC.490G23.2. Row and column represented time and survival rate, respectively. Low level and high level were based on the median of gene expression.

Figure 6.

DElncRNA-DEmRNA co-expression network. The ellipses and rhombuses represent the DEmRNAs and DElncRNAs, respectively. Red and green colors represent up- and downregulation, respectively. The top10 up/down-regulated DElncRNA/DEmRNA were with yellow border.

Figure 7.

Significant enrichment GO terms and KEGG pathways of DEmRNAs co-expressed with DElncRNAs in OSCC. (A) Biological process. (B) Molecular function. (C) Cellular component. (D) KEGG pathways. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; BP, biological process; FDR, false discovery rate; MF, molecular function; CC, cellular component.

3.4 Correlation analysis between the optimal diagnostic lncRNAs and clinical data

Correlation between the 15 lncRNAs and clinical data was analyzed, and the results were shown on the Box-plot (Fig. 5A) and survival curve (Fig. 5B). The Box-plot illustrated that the expression level of the 15 lncRNAs in normal controls were significant lower than which in OSCC, regardless the grades of OSCC. The survival analysis illustrated that the expression level of FOXD2.AS1 was significant association with the survival rate ( $P=$ 0.0275), which indicated that FOXD2.AS1 could be a prognostic indicator for OSCC.

3.5 Identification and functional analysis of co-expressed mRNAs of the optimal diagnostic lncRNAs

A total of 662 DEmRNAs co-expressed with the 15 DElncRNAs were obtained, resulting in 1333 co-expression pairs and the interacted network was represented in Fig. 6. The top three DElncRNAs that covered the most DEmRNAs were RP11.760H22.2 (degree $=$ 388), ADAMTS9.AS1 (degree $=$ 156) and CTC.297N7.9 (degree $=$ 125). Furthermore, FOXD2.AS1 correlated to survival rate was co- expressed with 19 DEmRNAs.

In addition, the 662 DEmRNAs were annotated by GO and KEGG analysis (Fig. 7). GO analysis showed that most DEmRNAs significantly enriched into the “signal transduction (FDR $=$ 1.08E-06)” in biological process, “cytoplasm (FDR $=$ 1.57E-14)” in cellular component and “protein binding (FDR $=$ 7.40E-07)” in molecular function. KEGG analysis showed that the “Focal adhesion (FDR $=$ 4.58E-14)” pathway enriched most DEmRNAs, followed by “ECM-receptor interaction (FDR $=$ 1.21E-18)” and “Pathways in cancer (FDR $=$ 3.52E-06)”.

4. Discussion

Machine learning and survival analysis indicated that FOXD2.AS1 could be a prognostic indicator for OSCC. FOXD2.AS1 was a lncRNA which had been made progress only in recent years [27]. Previous studies have shown that FOXD2.AS1 exhibited higher expression in various malignancies, including gastric [28], lung [29], bladder [30], colorectal [31], nasopharyngeal [32], esophageal [33], hepatocellular [34], thyroid [35] and skin cancer [36]. Multiple researches demonstrated that FOXD2.AS1 could active Wnt/ $\beta$ -catenin [29], EMT and Notch signaling pathway [37]. In addition, FOXD2.AS1 might promote the proliferation of cancer cell by competing endogenous RNA (ceRNA) manner, like binding with miR-185-5p participated in colorectal cancer progression [38]. Due to the few studies on the interaction between mRNAs and FOXD2.AS1, we identified 19 co-expressed DEmRNAs of FOXD2.AS1, large amounts of which were shown higher expression in patients with cancers. Among them, CA9, which was generally considered a marker of malignancy [39], was verified to express only in tumor cell of OSCC but not in normal tissues [40]. Meanwhile, ICAM5 was found highly expressed in head and neck squamous carcinoma tumorigenesis and perineural invasion, and FZD2 was reported to regulated cell proliferation and invasion in tongue squamous cell carcinoma [41, 42]. Previous studies indicated that these genes might function in cancers through Wnt, TGF and PI3K pathways [43]. Beyond FOXD2.AS1, 14 key lncRNAs associated with OSCC could also be the diagnostic indicators in OSCC regardless of tumor grades, which could be supported by several previous researches [44, 45, 46]. Notably, whether the identified diagnostic and prognostic markers were viable or not still need to be determined by clinical practices.

Regarding the functional annotation of co-expressed DEmRNAs of the 15 lncRNAs, “Focal adhesion” pathways and “ECM-receptor interaction” were found to be involved in various cancers. Studies in kidney cancer revealed that ECM-receptor interaction pathway correlated with cell migration and invasion and dysregulation of the focal adhesion pathway could also enhance the cell migration and invasion [47, 48].

5. Conclusion

In conclusion, we identified 15 diagnostic lncRNAs, verified the correlation of FOXD2.AS1 and survival rate, and revealed pathways of “Focal adhesion” and “ECM-receptor interaction” might be involved in OSCC.

Footnotes

Acknowledgments

This work was funded by Key Research and Development Projects of Sichuan Science and Technology Department (2018FZ0113), China Stomatological Association Western Medicine Stomatology Clinical Research Fund Project (CSA-W2016-01), Sichuan Science and Technology Support Program (2014SZ0038) and the National Natural Science Foundation of China (81600819).

Conflict of interest

The author declared there was no conflict of interest.

References

Durr

M.L.

and Wang

S.J.

, Oral cavity squamous cell carcinoma in never smokers: Analysis of clinicopathologic characteristics and survival, Am J Otolaryngol 34 (2013), 388–393.

Morgan

and Odell

E.W.

, Biopsy pathology of the oral tissues, Chapman & Hall Medical, 1997.

Silverman

, Jr, Demographics and occurrence of oral and pharyngeal cancers: The outcomes, the trends, the challenge, The Journal of the American Dental Association 132 (2001), 7S–11S.

Hawkins

R.J.

Wang

and Leake

J.L.

, Preventive health care, 1999 update: Prevention of oral cancer mortality, Journal-Canadian Dental Association 65 (1999), 617–617.

Shugars

D.C.

and Patton

L.L.

, Detecting, diagnosing, and preventing oral cancer, The Nurse Practitioner 22 (1997), 105, 109–110, 113–115 passim.

Wang

Liu

and Dong

, FusionCancer: A database of cancer fusion genes derived from RNA-seq data, Diagnostic Pathology 10 (2015), 131.

Yoshimura

Oka

and Harada

, Squamous cell carcinoma – Antigen for detection of squamous cell and mucoepidermoid carcinoma after primary treatment: A preliminary report, Journal of Oral and Maxillofacial Surgery 48 (1990), 1288–1292.

Silverman

N.A.

Alexander

J.C.

, Jr and Chretien

P.B.

, CEA levels in head and neck cancer, Cancer 37 (1976), 2204–2211.

Liu

S.A.

Tsai

W.C.

Wong

Y.K.

Lin

J.C.

Poon

C.K.

Chao

S.Y.

Hsiao

Y.L.

Chan

M.Y.

Cheng

C.S.

and Wang

C.C.

, Nutritional factors and survival of patients with oral cancer, Head & Neck: Journal for the Sciences and Specialties of the Head and Neck 28 (2006), 998–1007.

10.

Yamazaki

Chiba

Ishikawa

Satoh

Notani

K.-i.

Ohiro

Totsuka

Mizuno

and Kitagawa

, Serum p53 antibodies as a prognostic indicator in oral squamous cell carcinoma, Odontology 96 (2008), 32–37.

11.

SahebJamee

Eslami

AtarbashiMoghadam

and Sarafnejad

, Salivary concentration of TNF alpha, IL1 alpha, IL6, and IL8 in oral squamous cell carcinoma, Medicina Oral Patologia Oral y Cirugia Bucal 13 (2008), 292.

12.

Liao

P.-H.

Chang

Y.-C.

Huang

M.-F.

Tai

K.-W.

and Chou

M.-Y.

, Mutation of p53 gene codon 63 in saliva as a molecular marker for oral squamous cell carcinomas, Oral Oncology 36 (2000), 272–276.

13.

Vairaktaris

Serefoglou

Yapijakis

Stathopoulos

Vassiliou

Derka

Nkenke

Vylliotis

Ragos

and Neukam

F.W.

, The interleukin-1 beta gene polymorphism+ 3953 C/T is not associated with risk for oral cancer, Anticancer Research 27 (2007), 3981–3986.

14.

Vairaktaris

Yiannopoulos

Vylliotis

Yapijakis

Derka

Vassiliou

Nkenke

Serefoglou

Ragos

and Tsigris

, Strong association of interleukin-6-174 G

>

C promoter polymorphism with increased risk of oral cancer, SAGE Publications Sage UK: London, England, 2006.

15.

Vairaktaris

Yapijakis

Serefoglou

Derka

Vassiliou

Nkenke

Vylliotis

Wiltfang

Avgoustidis

and Critselis

, The interleukin-8 (-251A/T) polymorphism is associated with increased risk for oral squamous cell carcinoma, European Journal of Surgical Oncology (EJSO) 33 (2007), 504–507.

16.

Vairaktaris

Vylliotis

Spyridonodou

Derka

Vassiliou

NKENKE

Yapijakis

Serefoglou

Neukam

F.W.

and Patsouris

, A DNA polymorphism of stromal-derived factor-1 is associated with advanced stages of oral cancer, Anticancer research 28 (2008), 271–275.

17.

Yapijakis

Vairaktaris

Vassiliou

Vylliotis

Nkenke

Nixon

Derka

Spyridonidou

Vorris

and Neukam

, The low VEGF production allele of the+ 936C/T polymorphism is strongly associated with increased risk for oral cancer, Journal of Cancer Research and Clinical Oncology 133 (2007), 787–791.

18.

Franzmann

E.J.

Reategui

E.P.

Pedroso

Pernas

F.G.

Karakullukcu

B.M.

Carraway

K.L.

Hamilton

Singal

and Goodwin

W.J.

, Soluble CD44 is a potential marker for the early detection of head and neck cancer, Cancer Epidemiology and Prevention Biomarkers 16 (2007), 1348–1355.

19.

Mager

D.L.

Haffajee

A.D.

Devlin

P.M.

Norris

C.M.

Posner

M.R.

and Goodson

J.M.

, The salivary microbiota as a diagnostic indicator of oral cancer: A descriptive, non-randomized study of cancer-free and oral squamous cell carcinoma subjects, Journal of Translational Medicine 3 (2005), 27.

20.

Tang

Zhang

and Su

, Salivary lncRNA as a potential marker for oral squamous cell carcinoma diagnosis, Molecular Medicine Reports 7 (2013), 761–766.

21.

J.-Y.

Chung

H.-R.

Wang

D.-J.

Chang

W.-C.

Lee

S.-Y.

Lin

C.-T.

Yang

Y.-C.

and Yang

W.-C.V.

, Potential biomarkers in saliva for oral squamous cell carcinoma, Oral Oncology 46 (2010), 226–231.

22.

J.-S.

F.-J.

P.-Y.

K.-W.

C.-C.

and C.-J.

, Clinicopathological and prognostic significance of preoperative serum epidermal growth factor levels in patients with oral squamous cell carcinoma, International Journal of Oral and Maxillofacial Surgery.

23.

Liu

Lao

Liang

Zhang

Liao

and Liang

, Neck observation versus elective neck dissection in management of clinical T1/2N0 oral squamous cell carcinoma: A retrospective study of 232 patients, Chinese Journal of Cancer Research 29 (2017), 179–188.

24.

Hosni

Mcmullen

Huang

S.H.

Bayley

Bratman

S.V.

Cho

Giuliani

and Kim

, Lymph node ratio relationship to regional failure and distant metastases in oral cavity cancer, Radiotherapy & Oncology Journal of the European Society for Therapeutic Radiology & Oncology 124 (2017), 225.

25.

Mitchell

T.M.

, Does machine learning really work? AI magazine 18 (1997), 11–11.

26.

Libbrecht

M.W.

and Noble

W.S.

, Machine learning applications in genetics and genomics, Nature Reviews. Genetics 16 (2015), 321–332.

27.

Tai

and Wang

, Oncogenicity of lncRNA FOXD2-AS1 and its molecular mechanisms in human cancers, Pathology – Research and Practice 215 (2019), 843–848.

28.

T.-p.

Wang

W.-y.

Shuai

Zhao

Wang

Y.-f.

Xia

Chen

W.-m.

and Zhang

E.-b.

, Upregulation of the long noncoding RNA FOXD2-AS1 promotes carcinogenesis by epigenetically silencing EphB3 through EZH2 and LSD1, and predicts poor prognosis in gastric cancer, Oncogene 37 (2018), 5020–5036.

29.

Rong

Zhao

and Lu

, Highly expressed long non-coding RNA FOXD2-AS1 promotes non-small cell lung cancer progression via Wnt/β-catenin signaling, Biochemical and Biophysical Research Communications 484 (2017), 586–591.

30.

Zhang

J.-X.

Chen

Z.-H.

Chen

D.-L.

Tian

X.-P.

Wang

C.-Y.

Zhou

Z.-W.

Gao

Chen

and Zheng

Z.-S.

, LINC01410-miR-532-NCF2-NF-kB feedback loop promotes gastric cancer angiogenesis and metastasis, Oncogene 20 (2018).

31.

Yang

Duan

and Zhou

, Long non-coding RNA FOXD2-AS1 functions as a tumor promoter in colorectal cancer by regulating EMT and Notch signaling pathway, Eur Rev Med Pharmacol Sci 21 (2017), 3586–3591.

32.

Chen

Sun

Hua

Zeng

and Yang

, Long non-coding RNA FOXD2-AS1 aggravates nasopharyngeal carcinoma carcinogenesis by modulating miR-363-5p/S100A1 pathway, Gene 645 (2018), 76–84.

33.

Bao

Zhou

Zhang

and Diao

, Upregulation of the long noncoding RNA FOXD2-AS1 predicts poor prognosis in esophageal squamous cell carcinoma, Cancer Biomark 21 (2018), 527–533.

34.

Chang

Zhang

Zhou

Qiu

Wang

Chang

and Fan

, Long non-coding RNA FOXD2-AS1 plays an oncogenic role in hepatocellular carcinoma by targeting miR-206, Oncology Reports 40 (2018), 3625–3634.

35.

Zhang

Zhou

and Gao

, LncRNA FOXD2-AS1 accelerates the papillary thyroid cancer progression through regulating the miR-485-5p/KLK7 axis, Journal of Cellular Biochemistry 120 (2019), 7952–7961.

36.

Ren

Zhu

and Wu

, FOXD2-AS1 correlates with the malignant status and regulates cell proliferation, migration, and invasion in cutaneous melanoma, Journal of Cellular Biochemistry 120 (2019), 5417–5423.

37.

Yang

Duan

and Zhou

, Long non-coding RNA FOXD2-AS1 functions as a tumor promoter in colorectal cancer by regulating EMT and Notch signaling pathway.

38.

Zhu

Qiao

Zhou

Wang

and Zhou

, Long nonâ€coding RNA FOXD 2â€AS 1 contributes to colorectal cancer proliferation through its interaction with micro RNAâ€185â€5p, Cancer Science 109 (2018), 2235–2242.

39.

Chien

M.H.

Yang

J.S.

Chu

Y.H.

Lin

C.H.

Wei

L.H.

Yang

S.F.

and Lin

C.W.

, Impacts of CA9 gene polymorphisms and environmental factors on oral-cancer susceptibility and clinicopathologic characteristics in Taiwan, PLoS One 7 (2012), e51051.

40.

Choi

S.-W.

Kim

J.-Y.

Park

J.-Y.

Cha

I.-H.

Kim

and Lee

, Expression of carbonic anhydrase IX is associated with postoperative recurrence and poor prognosis in surgically treated oral squamous cell carcinoma, Human Pathology 39 (2008), 1317–1322.

41.

Maruya

S.I.

Myers

J.N.

Weber

R.S.

Rosenthal

D.I.

Lotan

and El-Naggar

A.K.

, ICAM-5 (telencephalin) gene expression in head and neck squamous carcinoma tumorigenesis and perineural invasion! Oral Oncology 41 (2005), 580–588.

42.

Huang

Luo

E.L.

Xie

Gan

R.H.

Ding

L.C.

B.H.

Zhao

Lin

L.S.

Zheng

D.L.

and Lu

Y.G.

, FZD2 regulates cell proliferation and invasion in tongue squamous cell carcinoma, Int J Biol Sci 15 (2019), 2330–2339.

43.

Wang

J.-K.

Wang

W.-J.

Cai

H.-Y.

B.-B.

Mai

Zhang

L.-J.

Y.-G.

Feng

S.-F.

and Miao

G.-Y.

, MFaP2 promotes epithelial-mesenchymal transition in gastric cancer cells by activating TgF-β/sMaD2/3 signaling pathway, Onco Targets and Therapy 11 (2018), 4001.

44.

Wang

F.-L.

Fei

Y.-H.

Wang

Y.-T.

Zhang

J.-E.

H.-Y.

and Zhang

, Aberrant expressed long non-coding RNAs in laryngeal squamous-cell carcinoma, American Journal of Otolaryngology 40 (2019), 615–625.

45.

Song

Pan

and Liu

, Functional analysis of lncRNAs based on competitive endogenous RNA in tongue squamous cell carcinoma, PeerJ 7 (2019), e6991.

46.

Rui

Feng-Qing

Lin-Li

De-Si

Yan

Jia-Rui

and Ming

, Comprehensive analysis of the whole coding and non-coding RNA transcriptome expression profiles and construction of the circRNA-lncRNA co-regulated ceRNA network in laryngeal squamous cell carcinoma, Functional & Integrative Genomics.

47.

Yamasaki

Seki

Yoshino

Itesako

Hidaka

Yamada

Tatarano

Yonezawa

Kinoshita

Nakagawa

and Enokida

, MicroRNA-218 inhibits cell migration and invasion in renal cell carcinoma through targeting caveolin-2 involved in focal adhesion pathway, J Urol 190 (2013), 1059–1068.

48.

Zhang

H.J.

Tao

Sheng

Rong

R.M.

and Zhu

T.Y.

, RETRACTED: Twist2 promotes kidney cancer cell proliferation and invasion via regulating ITGA6 and CD44 expression in the ECM-Receptor-Interaction pathway, Biomed Pharmacother 81 (2016), 453–459.

Identification of diagnostic and prognostic lncRNA biomarkers in oral squamous carcinoma by integrated analysis and machine learning

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS

Keywords

1. Introduction

2. Materials and methods

2.1 Eligible data of HNSCC screening in TCGA database

2.2 Identification of DEmRNAs and DElncRNAs in OSCC

2.3 Identification of optimal diagnostic lncRNA biomarkers for OSCC

2.5 Co-expression analysis of the optimal diagnostic lncRNAs and mRNAs

2.6 Functional analysis of co-expressed mRNAs of the optimal diagnostic lncRNAs

3. Results

3.1 Eligible data collection from TCGA database

3.3 Identification of the optimal diagnostic lncRNAs

Table 2 Differentially expressed lncRNAs between OSCC and normal tissues following reduced dimensions of the data

3.5 Identification and functional analysis of co-expressed mRNAs of the optimal diagnostic lncRNAs

4. Discussion

5. Conclusion

Footnotes

Acknowledgments

Conflict of interest

References

Table 2
Differentially expressed lncRNAs between OSCC and normal tissues following reduced dimensions of the data