Content uploaded by Fatma Hilal Yagin
Author content
All content in this area was uploaded by Fatma Hilal Yagin on Apr 22, 2022
Content may be subject to copyright.
196
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
MEDICAL RECORDS-International Medical Journal
Articial Intelligence-based Colon Cancer Prediction by
Identifying Genomic Biomarkers
Genomik Biyobelirteçleri Belirleyerek Yapay Zeka Tabanlı Kolon Kanseri
Tahmini
Nur Paksoy, Fatma Hilal Yagin
Malatya Fahri Kayahan Healthcare Center, Department of Family Medicine Physician, Malatya, Turkey
Inonu University, Faculty of Medicine, Department of Biostatistics and Medical Informatics, Malatya, Turkey
Copyright@Author(s) - Available online at www.dergipark.org.tr/tr/pub/medr
Content of this journal is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Received: 22.02.2022 Accepted: 28.04.2022
Corresponding Author: Fatma Hilal Yagin, Inonu University, Faculty of Medicine, Department of Biostatistics and
Medical Informatics, Malatya, Turkey, E-mail: hilal.yagin@inonu.edu.tr
Abstract
Aim: Colon cancer is the third most common type of cancer worldwide. Because of the poor prognosis and unclear preoperative
staging, genetic biomarkers have become more important in the diagnosis and treatment of the disease. In this study, we aimed to
determine the biomarker candidate genes for colon cancer and to develop a model that can predict colon cancer based on these
genes.
Material and Methods: In the study, a dataset containing the expression levels of 2000 genes from 62 different samples (22 healthy and
40 tumor tissues) obtained by the Princeton University Gene Expression Project and shared in the gshare database was used. Data
were summarized as mean ± standard deviation. Independent Samples T-Test was used for statistical analysis. The SMOTE method
was applied before the feature selection to eliminate the class imbalance problem in the dataset. The 13 most important genes that
may be associated with colon cancer were selected with the LASSO feature selection method. Random Forest (RF), Decision Tree (DT),
and Gaussian Naive Bayes methods were used in the modeling phase.
Results: All 13 genes selected by LASSO had a statistically signicant difference between normal and tumor samples. In the model
created with RF, all the accuracy, specicity, f1-score, sensitivity, negative and positive predictive values were calculated as 1. The RF
method offered the highest performance when compared to DT and Gaussian Naive Bayes.
Conclusion: In the study, we identied the genomic biomarkers of colon cancer and classied the disease with a high-performance
model. According to our results, it can be recommended to use the LASSO+RF approach when modeling high-dimensional microarray
data.
Keywords: Colon cancer, microarray, genomics, LASSO, random forest, decision tree, gaussian naive bayes
Öz
Amaç: Kolon kanseri dünya genelinde en sık görülen üçüncü kanser türüdür. Kötü prognoz ve net olmayan preoperatif evreleme
nedeniyle, hastalığın tanı ve tedavisinde genetik biyobelirteçler daha önemli hale gelmiştir. Bu çalışmada kolon kanseri için biyobelirteç
adayı genlerin belirlenmesi ve bu genlere dayalı olarak kolon kanserini başarılı bir şekilde tahmin eden bir modelin geliştirilmesi
amaçlanmıştır.
Materyal ve Metot: Çalışmada, Princeton Üniversitesi Gen Ekspresyon Projesi ile elde edilen ve gshare veri tabanında paylaşılan
62 farklı örnekten (22 sağlıklı ve 40 tümör dokusu) 2000 genin ekspresyon düzeylerini içeren bir veri seti kullanıldı. Veriler ortalama
± standart sapma olarak özetlendi. İstatistiksel analizler için bağımsız örneklerde T-testi kullanıldı. Veri setindeki sınıf dengesizliği
sorununu ortadan kaldırmak için öznitelik seçiminden önce SMOTE yöntemi uygulandı. Kolon kanseri ile ilişkili olabilecek en önemli
13 gen, LASSO öznitelik seçim yöntemi ile seçildi. Modelleme aşamasında Rastgele Orman (RF), Karar Ağacı (DT) ve Gauss naive
Bayes yöntemleri kullanıldı.
Bulgular: LASSO tarafından seçilen 13 genin tümü, normal ve tümör numuneleri arasında istatistiksel olarak anlamlı bir farka sahipti.
RF ile oluşturulan modelde doğruluk, seçicilik, f1-skor, duyarlılık, negatif ve pozitif prediktif değerlerinin tümü 1 olarak hesaplanmıştır.
DT ve Gaussian Naive Bayes ile karşılaştırıldığında RF yöntemi en yüksek performansı vermiştir.
Sonuç: Çalışmada kolon kanserinin genomik biyobelirteçlerini belirledik ve hastalığı yüksek performanslı bir model ile sınıflandırdık.
Elde ettiğimiz sonuçlara göre, yüksek boyutlu mikrodizi verilerinin modellenmesinde LASSO+RF yaklaşımının kullanılması önerilebilir.
Anahtar Kelimeler: Kolon kanseri, mikrodizi, genomik, LASSO, rastgele orman, karar ağacı, gaussian naive bayes
Research Article
197
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
INTRODUCTION
According to the World Health Organization, cancer is the
second leading cause of death after cardiovascular disease.
Colon cancer ranks 3rd in the world in terms of incidence
and is the 4th most common cancer. With the introduction
of screening programs in the USA in the last 30 years,
an improvement in cancer prognosis has been detected
thanks to early diagnosis, and this screening program
has been implemented in our country since 2009 (1,2).
Being able to perform postoperative staging and
determining prognosis by staging alone emphasizes
biomarkers and genetic evaluation in colon cancer. For
this reason, examination of colon cancer based on genetic
biomarkers is very important in the diagnosis and treatment
of the disease (3).
Microarray technology has allowed the simultaneous
measurement of thousands of gene expressions.
Identifying disease-related biomarker candidate
genes using microarray gene expression datasets and
distinguishing (classifying) disease samples from non-
disease samples has been an important research topic in
biomedicine and medicine. However, the resulting large-
scale datasets created many barriers to computational
techniques. The high dimensionality problem affects most
microarray gene expression datasets where dimensionality
is high (up to tens of thousands of genes) and sample size
is small (normally up to hundreds). Also, the high noise-to-
variability ratio of microarray trials adds to the difculties
(4).
Machine learning methods are frequently used to overcome
current challenges. Machine learning; it can be dened
as obtaining previously unknown, valid and applicable
information from data stacks through a dynamic process.
In this process, many techniques such as clustering,
data summarization, learning classication rules, nding
dependency networks, developing predictive models,
variability analysis and anomaly detection are used. With
machine learning, condential information is retrieved
in database systems comprising large data stacks. This
process is done using statistics, mathematical disciplines,
modeling techniques, database technology and various
computer programs (5,6).
Before constructing classication models in machine
learning with high-dimensional microarray datasets, it is an
important step to remove disease-related genes from the
dataset using trait (gene) selection methods. In this way,
both biomarker candidate genes can be selected and the
performance of the classication models to be created will
be improved (7).
In this study, we aimed to determine biomarker candidate
genes for colon cancer by using gene expression dataset
and to develop a classication model that can provide
clinical decision support to healthcare professionals.
MATERIAL AND METHOD
Dataset
In this study, an open source colon cancer gene expression
dataset obtained by Princeton University Gene Expression
Project and shared in gshare database (https://gshare.
com/articles/dataset/The_microarray_dataset_of_colon_
cancer_in_csv_format_/13658790/1) was used (8). The
dataset includes expression levels of 2000 genes from 62
different samples (22 healthy and 40 tumor tissues).
Statistical Evaluation
Data were summarized as mean ± standard deviation.
Compliance with the normal distribution was done with
the Kolmogorov-Smirnov test. Independent Samples T
Test was used for statistical analysis. Statistical tests
with a p value of less than 5% were considered signicant.
All statistical analyzes were performed using IBM SPSS
Statistics for Windows version 26.0 (New York, USA).
Data Preprocessing and Modeling
In datasets with class imbalance problem, most machine
learning techniques ignore minority class performance and
therefore underperform in minority class. One approach
to these datasets is to oversample the minority class and
is called the Synthetic Minority Oversampling Technique,
or SMOTE for short (9). In order to eliminate the class
imbalance problem in the colon cancer gene expression
dataset (22 normal and 40 tumor tissues), the SMOTE
method was applied before feature selection. In this way,
the number of samples in the groups, 40 normal and 40
tumor tissues, was equalized.
Afterwards, the 13 most important genes that may be
associated with colon cancer were selected with the
LASSO feature selection method. For the generalizability
of the model, 80% of the data set is divided as the training
set and 20% as the test set. Random Forest, Decision Trees
and Gaussian Naive Bayes classication methods were
used to predict colon cancer based on selected genes. The
performance of the models was evaluated with accuracy,
specicity, sensitivity, f1-score, negative predictive value
and positive predictive value.
LASSO Feature Selection
In 1996 the LASSO method was rst used by Robert
Tibshirani. Regularization and property selection are the
two main tasks of the method. The LASSO method puts
a constraint on the sum of the absolute values of model
parameters; the sum must be less than a xed value (upper
bound). To do this, the method implements a narrowing
(regularization) process in which regression variables
punish their coefcients, some of which reduce them to
zero. During the property selection process, variables that
still have a coefcient of zero after collapse are selected
for the model. This operation minimizes the prediction
error. In practice, the parameter that controls the power of
198
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
punishment, is of great importance. When large enough,
the dimensionality is can be reduced in this manner. The
larger the parameter, the more coefcients are reduced
to zero. There are many advantages to using the LASSO
method. First, it can provide very good forecast accuracy,
since the reduction and removal of coefcients can reduce
variance without a signicant increase in deviation. It
is especially useful when there are few observations
and many variables in the data set. LASSO also helps to
improve the interpretability of the model by eliminating
irrelevant variables that are not associated with the
response variable, so that the problem of overlearning can
also be addressed (10,11).
Random Forest
The Random Forests algorithm, a community learning
method, aims to increase the classication value by
generating multiple decision trees during the classication
process. Because it includes random sampling and
improved properties of techniques in community methods,
the RF method offers better generalizations and makes
more valid predictions than conventional machine learning
methods. The reasons for the precise estimates of the RF
method are that it gives low deviation and low correlation
between trees. The low amount of deviation is obtained as
a result of the creation of rather large trees. By creating
as many different trees as possible, a low correlation
structure is achieved. Individually created classication
and regression decision trees come together to form the
decision forest community. The decision trees here are
randomly selected subsets from the data set to which
they are connected. The results obtained during the
formation of the decision forest are combined to make
the latest prediction. For classication, trees each leaf
node is created to contain only members of one class. For
regression, trees continue to divide until a small number of
units remain in the leaf node (12).
Decision Trees
Decision trees (DT) consist of root nodes, branches and
leaves. The leaves in the decision trees are the places
where the classication occurs and the branches refer
to the result. The tree is created by the division variation
method from the root node to the leaf nodes. A decision
node can contain one or more branches. A decision tree
can consist of both categorical and numerical data. The
decision tree contains two basic process steps. These
operations are splitting and pruning operations (13). The
most important step when creating a DT is to decide which
attribute values to base it on and which branching to create.
In the knowledge gain and gain ratio approach that includes
entropy rules, all attributes at hand are tested subjectically
and the attribute with the highest knowledge gain is
selected for branching. DT are a classication method that
creates a model in the form of a tree structure consisting of
decision nodes and leaf nodes by classication, property,
and target. The decision tree algorithm is developed by
dividing the data set into smaller pieces (14,15).
Gaussian Naive Bayes
A simple structured classication based on conditional
probability, which is assumed to be equal and independent
of each other in the classication of all attributes based
on conditional probability. The classication process is
done by combining the effects of different attributes on
the result. Naive Bayes classies using statistical methods
and is an important algorithm in terms of performance. The
importance of qualications is considered equal in all. The
Gaussian Naive Bayes (GNB) classier is the Naive Bayes
method, which is created by assuming that the class label
is a Gaussian distribution on the given property values. GNB
assigns all data to the closest location. However, instead of
using Euclidean distance to calculate the distance between
them, it calculates by taking into account the distance from
the average and the class variance (16).
RESULTS
Table I contains descriptive statistics for 13 genes selected
by LASSO trait selection. When Table I is examined; all 13
genes selected by LASSO had a statistically signicant
difference between normal and tumor samples. Hsa.8125,
Hsa.2710, Hsa.8147, Hsa.36689, Hsa.31933, Hsa.1387
and Hsa.865 were expressed lower in tumor samples, while
Hsa.3306, Hsa.22762, Hsa.3016, Hsa.5392, Hsa.1410 and
Hsa.2928 were expressed higher in tumor samples.
Table 1. Descriptive statistics for selected genes
Gene Name Normal (Mean ± SD)Tumor (Mean ± SD) t value p-value
Hsa.8125 2.144 ± 0.496 1.444 ± 0.442 6.87 <0.001
Hsa.2710 1.289 ± 0.392 0.89 ± 0.359 5.3 <0.001
Hsa.8147 2.092 ± 0.799 0.725 ± 0.637 9.97 <0.001
Hsa.36689 0.741 ± 0.42 -0.01 ± 0.318 9.83 <0.001
Hsa.3306 0.289 ± 0.504 1.138 ± 0.482 -8.07 <0.001
Hsa.22762 -0.242 ± 0.564 0.337 ± 0.759 -4.05 0.003
Hsa.31933 -0.107 ± 0.263 -0.475 ± 0.377 5.24 <0.001
Hsa.3016 -0.222 ± 1.074 0.962 ± 1.049 -5.16 <0.001
Hsa.5392 -1.064 ± 0.655 -0.486 ± 0.486 -4.82 <0.001
Hsa.1410 -0.794 ± 0.73 0.002 ± 0.694 -5.53 <0.001
Hsa.2928 -1.312 ± 0.563 -0.526 ± 0.518 -6.98 <0.001
Hsa.1387 0.827 ± 0.648 0.017 ± 0.779 5.4 <0.001
Hsa.865 0.45 ± 0.393 0.06 ± 0.568 3.78 0.006
SD: Standard deviation
199
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
Table II presents the results of the performance measures
of the RF, DT, and GNB classication models. Specicity,
accuracy, f1-score, sensitivity, negative and positive
predictive value criteria obtained from the RF model
were all calculated as 1. That is, the RF model correctly
predicted all samples in the test set. From the DT model,
all performance measures were obtained as 0.9. Finally, in
the model created with the GNB method, the performance
measures were found to be accuracy 0.95, specicity 1, f1-
score 0.95, sensitivity 0.9, negative predictive value 0.9091,
and positive predictive value 1. The RF method offered the
highest performance compared to DT and GNB.
Table 2. Performance measures results for classication models
Metric Random Forest Gaussian Naive Bayes Decision Trees
Accuracy 1 0.95 0.9
Sensitivity 1 0.9 0.9
Specicity 1 1 0.9
PPV 1 1 0.9
NPV 1 0.9091 0.9
F1 score 1 0.95 0.9
PPV: Positive predictive value, NPV: Negative predictive value
DISCUSSION
Since knowing the biological functions of genes is useful
for knowing the origin, causes and treatment of many
diseases, studies in the eld of genomics have been on
the agenda of the scientic world for years. In addition to
their biological functions, the detection and relationships
of genes in the same biological pathway bring microarray
studies to the fore. Thanks to the detection of possibly
related genes, the detection and treatment of diseases
has become easier with the identication of gene clusters
(17). Based on this information, in the current study,
we developed a model that can predict the disease by
identifying the genes associated with colon cancer to
provide clinical decision support to physicians.
In this study, we used the LASSO feature selection method
to identify colon cancer-related genes. With the LASSO
method, Hsa.8125, Hsa.36689, Hsa.3306, Hsa.3016,
Hsa.8147, Hsa.2710, Hsa.22762, Hsa.31933, Hsa.5392,
Hsa.1410, Hsa.2928, Hsa.1387 and Hsa.865 genes may
be associated with colon cancer. Some of the biomarker
candidate genes we identied were in agreement with the
literature. Shaik et al. showed differential expression of
Hsa.8125, Hsa.36689 and Hsa.3306 genes in colon cancer
(genes1). In another study, Hsa.8125 and Hsa.3306 were
among 100 genes associated with colon cancer (18).
Hsa.8125; it is a gene that activates RNA binding activity,
is involved in nucleocytoplasmic transport, is located in the
endoplasmic reticulum, nucleus and perinuclear region of
the cytoplasm. Yan et al. showed that this gene, also known
as ANP32A, is overexpressed in colorectal cancer patients
and ANP32A levels are higher in poorly differentiated
tumors (19). Velmurugan et al. reported that this gene is
associated with lymph node metastasis (20).
When the relationship between the Hsa.36689 gene, whose
main task is guanylate cyclase activation in the colon, and
colon cancer was examined, Yang et al. identied this gene
among the top 5 most related genes (21). The Hsa.3016
and Hsa.8147 genes that we detected were also detected
as the other genes with the highest frequency in this study.
The Hsa.3306 gene is a gene that plays a role in cell
proliferation and is increased in cancer. In another study
examining the colon gene data set in the literature, it was
identied as one of the ten most closely related genes
among 2000 genes due to its association with colon
cancer. Among the genes detected in this study, Hsa.8125,
one of the genes we detected, is also included. It has been
shown that this gene, whose functions are important in the
construction of intestinal villi, increases in normal cells and
decreases in colon cancer cells (22).
Hsa.8147, also known as the desmin gene, is the gene
responsible for the production of desmin, a smooth
muscle-type intermediate lament protein expressed by
smooth muscle cells, but also in brotic tissue in wound
healing and tumor ‘desmoplastic’ stroma. Desmin also
surrounds the vasculature by being produced by pericytes
during angiogenesis in capillaries. It also plays a role in
angiogenesis in cancer tissue. Studies have shown an
increase in desmin expression in advanced cancer patients
(23). In a study conducted in patients with gallbladder
cancer, down-regulation of the desmin gene was detected
(24).
The Hsa.3016 gene, which we have observed to be
strongly associated with colon cancer, is one of the
genes responsible for coding the S-100P protein. S100
proteins are involved in many events such as regulation
of calcium homeostasis, cell proliferation, apoptosis, and
cell migration. The S100 protein family plays a role in many
stages of cancer formation and progression. S-100P acts
as an inducer of metastasis, overexpression of S-100P
increases the expression of S-100A6 and Cathepsin D,
which are involved in cellular invasion. Furthermore, S100P
promotes transendothelial migration of tumor cells (25).
The Hsa.2710 gene is one of the genes responsible for
making Fibulin-1, a secreted glycoprotein that is included
in the brillar extracellular matrix. It is involved in cell
adhesion and migration along protein bers within the
extracellular matrix (ECM). Considered to have a role in
cellular transformation and tumor invasion, it acts as a
tumor suppressor (26). In the study of Xu et al. , it was
shown that bulin downregulation is associated with
colorectal cancer (27).
Nucleolin; it is a multifunctional protein that is also found
in the nucleolus, nucleoplasm, and cytoplasm. Hsa.22762
is one of the genes involved in the synthesis of nucleolin.
200
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
It is involved in the regulation of translation and stability
of oncogenic mRNAs in the nucleoplasm. In our study, the
presence of this gene was found to be signicantly related
in colon cancer patients. It has also been shown in other
studies that nucleolin is overexpressed in many cancer
types such as stomach, pancreatic, breast, cervix, prostate
cancers, leukemias, melanomas and colorectal cancers
(28).
The Hsa.31933 gene, which we detected in our study, is
one of the genes that helps Autographa californica multiple
nuclear polyhedrosis virus (AcMNPV), which is from the
Baculovirus family, to successfully initiate the expression
of viral genes by preparing the host environment and
controlling the subsequent viral gene expression like other
DNA viruses to infect their hosts. Viral genes, which are
expressed immediately after infection, play a critical role
in the early infection process; Hsa.31933 (Immediate-
Early Regulatory Protein IE-N gene) is one of these genes.
AcMNPV has been studied as a gene therapy vector. In
a study by Ono et al., they determined AcMNPV induces
antitumor acquired immunity; they showed AcMNPV can
act as an effective immune-inducing virus and eukaryotic
expression vector for gene carrier and has the potential to
be a tumor therapy agent (29).
In another study, recombinant DNA obtained with this virus
enabled the production of a natural antigen associated
with carcinoma in mice (30). Although there are no studies
related to this virus DNA in colon cancer yet, the data in our
study showed that there is a strong relationship between
colon cancer and this gene. We think that meaningful
results can be obtained as a result of the use of AcMNPV
as a vector with more comprehensive studies on the
treatment of colon cancer.
The Hsa. 5392 gene is also known as ribosomal protein
L24 (RPL24). It is one of the genes responsible for the
expression of ribosomal proteins. It encodes the ribosomal
protein L24, a homolog of the cytosolic RPL24 found in
higher eukaryotes. Studies have been conducted on the
overexpression of a number of ribosomal protein genes in
human tumors and their contribution to tumorigenesis (31).
Hsa.1410 is the gene responsible for the synthesis of
the eukaryotic translation initiation factor eIF-2. The
role of protein synthesis changes is important in cancer
development and progression. Studies show that ribosomal
protein synthesis plays a direct role during tumor initiation.
The translation initiation process is the rate-limiting step of
protein synthesis in eukaryotes, and a group of eukaryotic
translation initiation factors (eIFs) are involved. In previous
studies, it has been shown that a signicant increase in eIF3
subunits, eIF3A, eIF3B and eIF3M overexpression, which is
one of the translation initiation factors, in colorectal cancer
patients, and eIF4 subunits, of which eIF3C is an oncogene,
are also increased in cancer cells (32).
In studies, eIF2a expression was described as transiently
increased in normal cells, whereas constitutive
overexpression indicated tumor initiation and progression.
Golob-Schwarzl et al., they also showed that eIF2 is
overexpressed in colorectal cancers (32).
Among the genes we determined, Hsa.2928 is the mRNA
gene responsible for the expression of P-cadherin.
Cadherins are calcium-dependent cell adhesion proteins
that provide cell architecture and integrity, and their
degradation is often associated with human cancer (33).
Neo-expression or up-regulation of placental cadherin
(P-cadherin) has been reported in a variety of carcinomas,
including colorectal and bladder carcinomas (34).
The Hsa.1387 Human 11 beta-hydroxysteroid
dehydrogenase type II mRNA gene is a gene that has
a strong association with colon cancer and has been
found to be associated with colon carcinomas. 11 beta-
Hydroxysteroid dehydrogenase type II enzyme (11 beta
HSD2), which is also located in the colon, which has an
important role in water and electrolyte homeostasis, gives
specicity to the mineralocorticoid receptor (35).
MAP kinases, also known as (ERKs) encoded by the Hsa.865
(ERK-1, M84490) gene, are regulated by extracellular
signaling and act in a signal cascade that regulates various
cellular processes such as proliferation, differentiation
and cell cycle through the action of extracellular signals.
The tumor suppressor pathway is stimulated by ERK-1
phosphorylation (36). The relationship between colon
cancer and ERK-1 has been shown in many studies (37-39).
In our study, we showed its relationship with colon cancer.
In a similar study using the same data set in the literature,
PCA and PLS feature extraction methods were applied
and then they classied colon cancer with the support
vector machine method with an accuracy of 0.9516
(40). In another study, they found that the combined use
of PSO and SVM outperformed the model created with
only the SVM algorithm in terms of accuracy (0.94) and
performance, and was faster in terms of time analysis
(41). In the current study, three models were created using
RF, DT and GNB classiers based on biomarker candidate
genes determined by LASSO feature selection method.
According to the performance criteria obtained, the LASSO
+ RF model showed the best performance by correctly
classifying all samples.
CONCLUSION
In conclusion, this study identied genomic biomarkers
of colon cancer and classied the disease with a high-
performance model. According to the results obtained,
the LASSO method gave results compatible with the
literature while determining the genomic biomarkers.
For this reason, genes selected with LASSO can provide
clinical decision support to physicians in the diagnosis and
treatment of colon cancer. In addition, it can be suggested
that the LASSO+RF approach be used in modeling high-
dimensional data in medicine.
Financial disclosures: The authors declared that this study
hasn’t received no nancial support.
201
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
Conflict of Interest: The authors declare that they have no
competing interest.
Ethical approval: Ethics committee approval is not required
in this study.
REFERENCES
1. Globocan W. Estimated cancer incidence, mortality and
prevalence worldwide in 2012. Int Agency Res Cancer. 2012.
2. Labianca R, Beretta G, Gatta G, et al. Colon cancer. Critical
Reviews Oncology Hematology. 2004;51:145-70.
3. Loboda A, Nebozhyn MV, Watters JW, et al. EMT is the
dominant program in human colon cancer. BMC Med
Genomics. 2011;4:1-10.
4. Xu C, Meng LB, Duan YC, et al. Screening and identication of
biomarkers for systemic sclerosis via microarray technology.
Int J Molecular Med. 2019;44:1753-70.
5. Ahmad MA, Eckert C, Teredesai A. Interpretable machine
learning in healthcare. Proceedings of the 2018 ACM
international conference on bioinformatics, Computational
Biology Health Informatics. 2018
6. Yagin FH, Yagin B, Arslan AK, Çolak C. Comparison of
Performances of Associative Classication Methods for
Cervical Cancer Prediction: Observational Study. Turkey
Clinics J Biostatistics. 2021;13:13:266-72.
7. Khaire UM, Dhanalakshmi R. High-dimensional microarray
dataset classication using an improved adam optimizer
(iAdam). J Ambient Intelligence Humanized Computing.
2020;11:5187-204.
8. Hameed SS, Hassan R, Hassan WH, et al. HDG-select: A novel
GUI based application for gene selection and classication
in high dimensional datasets. PloS One. 2021;16:e0246039.
9. Mulla GA, Demir Y, Hassan M. Combination of PCA
with SMOTE Oversampling for Classication of High-
Dimensional Imbalanced Data. Bitlis Eren University Science
and Technology Journal. 2021;10:858-69.
10. Güçkiran K, Cantürk İ, Özyilmaz L. DNA microarray gene
expression data classication using SVM, MLP, and RF
with feature selection methods relief and LASSO. Journal
of Suleyman Demirel University Institute of Science and
Technology. 2019;23:126-32.
11. Akyol K, Bayir Ş, Baha Ş. Importance of Attribute Selection
for Parkinson Disease. Academic Platform J Engineering
Sci. 2020;8:175-80.
12. Yilmaz R, Yagin FH. Early detection of coronary heart
disease based on machine learning methods. Med Records.
2022;4:1-6.
13. Secgin Y, Oner Z, Turan MK, Oner S. Gender prediction with
parameters obtained from pelvis computed tomography
images and decision tree algorithm. Med Science.
2021;10:356-61
14. Doğan Ş, Türkoğlu İ. Hypothyroidi and hyperthyroidi
detection from thyroid hormone parameters by using
decision trees. Fırat University Journal of Oriental Studies.
2007;5:163-9.
15. Pulat M, Kocakoç ID. Machine Learning and Decision in
Turkey. Bibliometric Analysis of Published Theses in the
Field of Trees. Journal of Management and Economics.
2021;28:287-308.
16. Kamel H, Abdulah D, Al-Tuwaijari JM. Cancer classication
using gaussian naive bayes algorithm. 2019 Int Engineering
Conference (IEC); 2019:36:165-5.
17. Quackenbush J. Microarray analysis and tumor
classication. New England J Med. 2006;354:2463-72.
18. Jose A. Gene selection by 1-d discrete wavelet transform
for classifying cancer samples using dna microarray date.
Ph.D. thesis, University of Akron, 2009.
19. Yan W, Bai Z, Wang J, et al. ANP32A modulates cell growth
by regulating p38 and Akt activity in colorectal cancer.
Oncology Reports. 2017;38:1605-12.
20. Velmurugan BK, Yeh K-T, Lee C-H, et al. Acidic leucine-
rich nuclear phosphoprotein-32A (ANP32A) association
with lymph node metastasis predicts poor survival in
oral squamous cell carcinoma patients. Oncotarget.
2016;7:10879.
21. Liu Q, Tan Y, Huang T, et al. TF-centered downstream gene
set enrichment analysis: Inference of causal regulators by
integrating TF-DNA interactions and protein post-translational
modications information. BMC Bioinformatics. 2010;11:1-
17.
22. Mora JAM, Ordoñez FM, Bonilla DA. Improvement of k-means
clustering algorithm performance in gene expression data
analysis through pre-processing with principal component
analysis and boosting. 2017;3:53-9.
23. Arentz G, Chataway T, Price TJ, et al. Desmin expression in
colorectal cancer stroma correlates with advanced stage
disease and marks angiogenic microvessels. Clinical
Proteomics. 2011;8:1-13.
24. Bhunia S, Barbhuiya MA, Gupta S, et al. Epigenetic
downregulation of desmin in gall bladder cancer reveals
its potential role in disease progression. Indian J Med
Research. 2020;151:311.
25. Chen H, Xu C, Qing’e Jin ZL. S100 protein family in human
cancer. Am J Cancer Res. 2014;4:89.
26. Twal WO, Czirok A, Hegedus B, et al. Fibulin-1 suppression
of bronectin-regulated cell adhesion and motility. J Cell Sci.
2001;114:4587-98.
27. Xu Z, Chen H, Liu D, Huo J. Fibulin-1 is downregulated through
promoter hypermethylation in colorectal cancer: a CONSORT
study. Med (Baltimore). 2015;94.e663
28. Tong X, Mirzoeva S, Veliceasa D, et al. Chemopreventive
apigenin controls UVB-induced cutaneous proliferation
and angiogenesis through HuR and thrombospondin-1.
Oncotarget. 2014;5:11413.
29. Ono C, Sato M, Taka H, et al. Tightly regulated expression
of Autographa californica multicapsid nucleopolyhedrovirus
immediate early genes emerges from their interactions and
possible collective behaviors. Plos One. 2015;10:e0119580.
30. Strassburg CP, Kasai Y, Seng BA, et al. Baculovirus
recombinant expressing a secreted form of a transmembrane
carcinoma-associated antigen. Cancer Res. 1992;52:815-21.
31. Loging WT, Reisman D. Elevated expression of ribosomal
202
Med Records 2022;4(2):196-202
DOI: 10.37990/medr.1077024
protein genes L37, RPP-1, and S2 in the presence of mutant
p53. Cancer Epidemiology and Prevention Biomarkers.
1999;8:1011-6.
32. Golob-Schwarzl N, Schweiger C, Koller C, et al. Separation
of low and high grade colon and rectum carcinoma by
eukaryotic translation initiation factors 1, 5 and 6. Oncotarget.
2017;8:101224.
33. Oliveira P, Sanges R, Huntsman D, et al. Characterization
of the intronic portion of cadherin superfamily members,
common cancer orchestrators. European J Human Genetics.
2012;20:878-83.
34. Van Marck V, Stove C, Jacobs K, et al. Pcadherin in adhesion
and invasion: Opposite roles in colon and bladder carcinoma.
Int J Cancer. 2011;128:1031-44.
35. Takahashi K, Sasano H, Fukushima K, et al. 11 beta-
hydroxysteroid dehydrogenase type II in human colon: a
new marker of fetal development and differentiation in
neoplasms. Anticancer Res. 1998;18:3381-8.
36. Baba Y, Nosho K, Shima K, et al. Prognostic signicance
of AMP-activated protein kinase expression and modifying
effect of MAPK3/1 in colorectl cancer. British J Cancer.
2010;103:1025-33.
37. Esteve-Puig R, Canals F, Colome N, et al. Uncoupling of the
LKB1-AMPKα energy sensor pathway by growth factors and
oncogenic BRAFV600E. PloS One. 2009;4:e4771.
38. Zheng B, Jeong JH, Asara JM, et al. Oncogenic B-RAF
negatively regulates the tumor suppressor LKB1 to promote
melanoma cell proliferation. Molecular Cell. 2009;33:237-47.
39. Kim MJ, Park IJ, Yun H, et al. AMP-activated protein kinase
antagonizes pro-apoptotic extracellular signal-regulated
kinase activation by inducing dual-specicity protein
phosphatases in response to glucose deprivation in HCT116
carcinoma. J Bio Chemistry. 2010;285:14617-27.
40. Arowolo MO, Isiaka RM, Abdulsalam SO, et al. A comparative
analysis of feature extraction methods for classifying colon
cancer microarray data. EAI Endorsed Transactions Scalable
Information Systems. 2017;4:1-6.
41. Al Rajab M, Lu J, Xu Q. Examining applying high performance
genetic data feature selection and classication algorithms
for colon cancer diagnosis. Computer Methods Programs
Bio Med. 2017;146:11-24.
