=Paper=
{{Paper
|id=Vol-2488/paper23
|storemode=property
|title=Exploratory Analysis of Neuroblastoma Data Genes Expressions Based on Bioconductor Package Tools
|pdfUrl=https://ceur-ws.org/Vol-2488/paper23.pdf
|volume=Vol-2488
|authors=Sergii Babichev,Bohdan Durnyak,Vsevolod Senkivskyy,Oleksandr Sorochynskyi,Mykhailo Kliap,Orest Khamula
|dblpUrl=https://dblp.org/rec/conf/iddm/BabichevDSSKK19a
}}
==Exploratory Analysis of Neuroblastoma Data Genes Expressions Based on Bioconductor Package Tools ==
https://ceur-ws.org/Vol-2488/paper23.pdf
Exploratory Analysis of Neuroblastoma Data Genes
Expressions Based on Bioconductor Package Tools
Sergii Babichev1,2[0000-0001-6797-1467], Bohdan Durnyak2[0000-0003-1526-9005],
Vsevolod Senkivskyy2[0000-0002-4510-540X], Oleksandr Sorochynskyi2[0000-0003-0964-2598],
Mykhailo Kliap3[0000-0003-1933-6148] and Orest Khamula2[0000-0003-0926-9156]
1Jan Evangelista Purkyne University in Usti nad Labem, Usti nad Labem, Czech Republic
sergii.babichev@ujep.cz
2Ukrainian Academy of Printing, Lviv, Ukraine
durnyak@uad.lviv.ua, senk.vm@gmail.com, somsoroka@gmail.com
3Uzhhorod National University, Uzhhorod, Ukraine
mihaylo.klyap@uzhnu.edu.ua
Abstract. The technique of gene expression profiles exploratory analysis on the
basis of the use of Bioconductor package tools is presented in the paper.
Applying this method allows forming the matrix of genes expression for further
gene regulatory networks reconstruction and simulation of the obtained model.
The gene expression profiles of human neuroblastoma cells obtained using high
throughput RNA-sequencing technique have been used as the experimental data
to evaluate the effectiveness of the appropriate step implementation. Applying
of the proposed technique involved removing low expressed genes at the first
step. The number of genes was reduced from 53186 to 7435 at this step. The
filtered gene expression profiles normalizing was considered at the next step.
The quality of data normalizing was evaluated by both various graphic tools and
using quantitative criterion, which was calculated based on the cluster analysis
for the samples which were previously distributed into clusters.
Keywords: RNA sequencing analysis, gene expression profiles, Bioconductor,
reducing, normalizing, exploratory analysis, clustering, clustering quality
criterion
1 Introduction
Development of technique of gene expression data processing in order to allocate high
expressed genes, which allows us to distinguish investigated samples, is one of the
current directions of modern bioinformatics. Implementation of this technique create
the conditions to reconstruct gene regulatory networks with high level of sensitivity.
The further simulation of the reconstructed gene regulatory networks can allow us to
better understand the character of genes interconnection and, as a result, it can help us
also to understand the ways of influence to the appropriate key genes in order to
change the expressions of genes of the network according to goal of the current task.
Two techniques are actual to form the array of gene expression
Copyright © 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution
4.0 International (CC BY 4.0)
2019 IDDM Workshops.
nowadays: DNA-microchip technique [1,2] and RNA-molecules sequencing method
[3,4]. DNA-microchip technique is significantly cheaper in comparison with RNA-
molecules sequencing method. As a result of this technique applying, we obtain the
matrix of light intensities. Then, four stages should be implemented to transform the
light intensities matrix into matrix of genes expression: background correction,
normalization, PM correction and summarization. Each of the stage can be
implemented by different ways [5–7]. This fact decreases the quality of obtained
matrix of genes expression.
Applying RNA-molecules sequencing method allows obtaining the number of
investigated genes for studied samples directly. In this reason, this method more exact
in comparison with DNA-microchip technique. The number of genes determines the
level of this gene activity or its expression. At the next step it is necessary to remove
non-expressed genes for all samples and gene with low level of expression. At this
stage it is appear the problem identification of boundary value which allows dividing
genes to lowly-expressed and highly-expressed. Moreover, the matrix of counts of
genes is not suitable for the following processing. Thus, initially, the data should be
normalized. This step assumes transformation the counts values into the same suitable
range. There are various normalized methods to process gene expression values.
However, it should be noted, that the task of objective selection of appropriate
normalizing method based on the quantitative criteria has not effective solution
nowadays. This fact indicates the actuality of the research.
2 Formal Problem Statement
A block chart of procedure to process the experimental data which are obtained by
RNA-molecules sequencing technique is presented in fig. 1. The studied dataset is
presented as a matrix of counts, values of which are determined the number of genes
corresponding to appropriate sample.
Fig. 1. A step-by-step procedure to transform a matrix of counts to the matrix of highly-
expressed informative genes
One of the most important steps of this procedure implementation is the data
normalizing. The normalized values of gene expression profiles should have the equal
ranges and their norms should be distinguish minimally between each other.
Moreover, the values of genes expression should allow us to identified the samples
which belong to various clusters. Considering hereinbefore, evaluation of quality of
gene expression profiles normalizing will be performed visually by analysis of both
box plot and kernel density plot, and based on quantitative criterion.
The main problem, which is solved within the framework of the current research,
consists in comparison analysis of various normalization techniques of gene
expression values using different normalizing quality criteria.
3 Literature Review
Various techniques have been proposed over last years to pre-process the results of
RNA-molecules sequencing experiments [8-16]. These tools are differed between
each other by types and thresholds which are used to process the counts of genes and
by algorithms which are used for filtering and alignment of the investigated values. It
should be noted that the choice of the alignment algorithm has very great influence to
the evaluation accuracy of RNA molecules abundance in the sequenced samples. In
this reason, testing different tools for processing of data of RNA-molecules
sequencing experiments can help us to choose the best technique for current type of
data.
After implementation of the alignment procedure, it is necessary to normalize the
recovered values of miRNA counts for purpose of removing variations in the data
which have not biological origins and, as a result, can influence to the ranges of
measured values change. Correct applying the normalizing technique allows
minimizing the experimental and the technical bias without noise introduce. In
[17,18] the authors have proposed several normalizing techniques for data of RNA-
molecules sequencing experiments. As a result of comparison of different
normalization methods effectiveness, the conflicting results were obtained in these
works. So, the authors in [18] proposed using the locally weighted linear regression
and quantile normalizing techniques. At the same time, they were discouraging
against the use of trimmed mean of M values (TMM technique). The obtained results
were validated based on the use of polymerase chain reaction (qPCR). In [17] the
authors proposed the opposite, to use against quantile normalization the TMM
method. The simulation results were used to confirm these findings. An assessment of
the relative effectiveness of various pre-processing techniques in terms of statistical
criteria, bias, sensitivity and specificity in order to detect the differential expressed
genes can be achieved on the basis of complex implementation of both qualitative and
quantitative normalizing quality criteria using current techniques of data processing
[19–26].
The aim of the paper is exploratory analysis of various technique of data from
RNA-molecules sequencing experiment processing based on the complex use of
Bioconductor package tools and various quality criteria to estimate the effectiveness
of the data processing.
4 Materials and Methods
4.1 Data Set
We used the dataset GSE129336 generated from Gene Expression Omnibus (GEO)
database [27] as the experimental data during the simulation process. The data
contains the results of expression profiling by high throughput sequencing in human
SH-SY5Y neuroblastoma cells [28]. The transcriptomic responses to Mn dose
(0,1,5,10,50,100 μM MnCl2 for 5 h) in the investigated cells with three biological
replicates per Mn treatment were examined during the experiment performing. Thus,
the examined samples can be divided into six clusters considering the Mn dose. Each
of the clusters contains three samples. This fact can be used to calculate one of the
criteria to estimate the quality of gene expression values processing. Each of the
samples contained 53186 of genes. So, the initial dataset contained 53186 of genes or
rows and 18 of columns or samples. The early analysis has shown, that there were
27838 non-expressed genes (zero for all samples). Of course, these genes can be
removed from the data at the first step. Moreover, the lowly-expressed genes for all
samples can be removed from the data too. The search of the thresholding value to
remove lowly-expressed genes is one of the solved tasks within the framework of this
research.
4.2 Removing Lowly-expressed Genes
As was noted in the section 4.1, the studied dataset contains 53186 of genes.
However, 27838 of genes are non-expressed for all samples (the count value is zero).
Thus, the number of the expressed genes can be changed from 53186 to 25348 of
genes.
At the next step, it is necessary to remove lowly expressed genes considering the
appropriate thresholding value. The initial values of the counts of genes are not
suitable for solve this task since the range of the genes count value change is very
large (in the case of our dataset this range is changed from o to 47434890). In this
case it is necessary to transform the count value scale into other, more suitable scale.
To solve this tack, Bioconductor package contains cpm() function which allows
transforming the counts values into count-per-million values as follows:
xij
x'ij n 106 (1)
xij
i 1
where n is number of rows, xij is value in i-th row and j-th column. Applying this
function allows us to obtain the new, more suitable, range of the data values change
(from 0 to 380367.8).
The main idea for lowly-expressed genes removing is the following: the use a
nominal thresholding of 1 cpm value (this value is corresponded to 0 log2(cpm) value)
allows dividing the genes into two groups (expressed and unexpressed). If value of
gene expression is more than this threshold, the gene is identified as expressed.
Otherwise, the gene is identified as unexpressed. Considering the number of samples
in the clusters we can suppose that the genes should be expressed in at least one
cluster (three samples) for the further analysis.
4.3 Normalizing Gene Expression Profiles
The following normalizing techniques were evaluated during the simulation process:
1) lcpm; 2) TMM; 3) TMMwsp; 4) RLE; 5) upper quartile scaling. Brief describing
each of these techniques is presented below.
1. lcpm – the simplest normalizing technique, the log2(cpm) values are calculated
during this technique implementation.
2. TMM – trimmed mean of M is the normalizing technique by total count of
scaling. The counts quantity for an appropriate target for all samples is estimated
during TMM technique implementation. If an expression value is identified in the
same proportion for all samples, this gene is identified as non-differentially expressed.
It should be noted that this technique does not allow considering the potentially
different RNA molecules which are presented in the samples. Applying this method
allows us to calculate a linear scaling index for appropriate sample considering
weighted average after transforming the data using log fold-changes (M) relative to
the absolute intensity in the reference sample (A) [29].
3. TMMwsp – TMM with singleton pairing. This technique is a variant of TMM,
in which the data with a high proportion of zeros are processed. Implementation of the
TMM method assumes that the genes which have zero value in either library are
ignored when pairs of libraries are compared between each other. As opposed to
TMM method, implementation of the TMMwsp technique assumes that the positive
counts from such genes are reused to increase the quantity of features which are used
to compare the libraries. The singleton positive counts are paired up between the
libraries in decreasing order of size and then a slightly modified TMM method is
applied to the re-ordered libraries.
4. RLE – relative log expression technique. Implementation of this method
assumes that the median library is calculated from the geometric average of all
columns and the median ratio of each sample to the median library is used as the scale
factor.
5. Upper quartile scaling – is the upper-quartile normalizing technique, in which
the scale factors are calculated from the 75% quantile of the counts for each of the
libraries, after removing genes that are zero in all libraries.
4.4 Quantitative Criterion to Estimate the Quality of Data Normalizing
The main idea to evaluate the quality of data normalizing is the following: as we
noted hereinbefore, the samples can be divided into six clusters considering the dose
of Mn. Each of the clusters in this case contains three samples. It is naturally that
informativity of gene expression profiles is determined by their ability to distinct the
samples in different clusters. Thus, the quality of data normalizing can be estimated
based on clustering quality criterion which should consider the samples distribution
within clusters and the clusters distribution in the feature space. Considering the high
dimension of the studied vectors, the correlation metric should be used to estimate the
proximity level between the vectors. This quality criterion of the samples and clusters
grouping was calculated as multiplicative combination of Calinski-Harabasz criterion
and WB-index [30,31]:
Nc ( Nc 1) QCW 2
QNC (2)
( N s Nc ) QCB 2
where: Nc is the clusters quantity; Ns is the samples quantity; QCW is an average
distance from samples to centers of the clusters where these samples are allocated;
QCB is an average distance between clusters’ centers. It should be noted that
minimum value of the criterion (2) corresponds the best normalizing technique.
5 Experiments, Results and Discussions
Fig. 2 presents the results of lowly expressed genes reducing in accordance with
technique which are described in 4.2. To increase the charts informativity the data
preliminarily were transformed using log2(cpm) function. The number of genes was
reduced at this step from 25348 to 7435. The analysis of the obtained in fig. 2
diagrams allows concluding that level of genes expression informativity significantly
increased due to remove lowly expressed values. The same conclusion can be done
based on the box plots analysis (See Fig. 3). In the case of filtered data use the values
of gene expressions for all samples are distributed more evenly and they are shifted to
the side of larger values.
Fig. 2. Density plots of non-filtered and filtered gene expression values distribution for
neuroblastoma data samples
The next step of the data preprocessing is their normalizing. Fig. 4 shows the chart
of the clustering quality criterion (2) versus the normalizing method. To calculate this
criterion values the data previously were divided into clusters considering the Mn
dose. It should be noted that in the case of non-normalized filtered data the value of
this criterion was 100,05.
Fig. 3. Box plots of non-filtered and filtered gene expression values distribution for
neuroblastoma data samples
Fig. 4. Dot plot of the quality criterion versus the normalizing method
The obtained results analysis allows concluding that normalizing process
significantly increases the quality of the data in terms of the quality criterion (2). The
value of this criterion for non-normalizing data 100,05 decreases more than 10 time.
Comparison analysis of various normalizing methods has shown that the easiest lcpm
method is showed the worst results in comparison with other methods. The difference
between methods TMM, TMMwsp, RLE and Upper quartile scaling is very small,
however, the value of the criterion (2) achieved the minimum one in the case of
Upper quartile scaling method apply. This fact indicates the reasonable of this method
use for current type of data normalizing.
At the next step it is necessary to remove heteroscedasticity from the data. The
analysis of the normalized data has shown that in the case of RNA-seq data use, the
variance values are not depend on the mean values. Methods that counts of the model
with the use of Negative Binomial distribution are based on a quadratic mean-
variance relationship. In limma package of R software, linear modelling is performed
using the normalized values. In this case the data should be normally distributed and
the mean-variance relationship is evaluated with the use of precision weights
calculated by the voom() function. Fig. 5 presents the results of this step
implementation.
Fig. 5. Visualization of the heteroscedasticity removing from the data
The left chart in the Fig.5 shows the mean-variance relationship of normalized
gene expression values. Usually, the voom-plot shows a decreasing trend between the
means and variances which are appeared due to an existence of both the technical
incorrectness during the sequencing experiment performing and the biological
variation among the replicate samples from various cell samples. Typically, the
results of the experiments with high level of biological variation are presented as a
flatter trends. The variance values in this case are not significantly changed for high
expression values (right chart in Fig. 5). And otherwise, experiments including data
with low biological variation usually have tend to sharp decreasing the variance
values. Moreover, the voom-plot allows us to visual evaluate the quality of gene
expression filtration process. If filtration process of lowly-expressed genes is
insufficient, then, the variance values should be decreased at the low end of the
expression scale due to very small gene expression values.
In order to visual summarize the results for all genes in obtained groups, we create
a mean-difference plots using the plotMD function of limma package. These plots
allow us to display log-Fold-change values from the linear model which can be fitted
against the average of log-expression values. These charts allow us to identify
differentially expressed genes. The charts are shown in Fig. 6.
Fig. 6. Mean-difference plots of gene expression profiles for investigated groups
Result of the visual analysis of the obtained diagrams allows concluding the
greatest number of genes in investigated groups have high level of differentially
expression (black color or number 1). It means that these genes are informative to
distinct the samples for the further processing. However, the data contains some
quantity of lowly-expressed genes (red color or zero number). It is means that these
data need the following processing for purpose of non-informative genes reducing in
terms of various quantitative criteria.
Fig. 7 shows the box charts of the processed gene expression profiles. The
samples previously were reordered considering Mn dose from 0 to 100 μM. Analysis
of character of gene expressions distribution allows us to conclude about correctness
of data preprocessing step implementation. The values of gene expression have the
same and not so much ranges, all genes are enough highly-expressed for all of the
samples. However, it should be noted, that there is some quantity of lowly-expressed
genes (for example, in Mn_1 sample). This fact indicates about necessity the further
data processing on the basis of the use of current techniques of complex data
processing.
Fig. 7. Results of gene expression profiles of neuroblastoma data processing
6 Conclusions
This paper presents the research results about processing of RNA-molecules
sequencing experiments. The dataset GSE129336 generated from Gene Expression
Omnibus database was used as the experimental one. This data contains the results of
expression profiling by high throughput sequencing in human SH-SY5Y
neuroblastoma cells. The initial data matrix contained counts of expressed genes for
studied samples. At the first step, we have removed lowly-expressed genes. The
number of genes was changed from 53186 to 7435. Then, we have compared various
normalizing technique using clustering quality criterion as the main criterion of
appropriate normalizing method effectiveness estimation. At the next steps we have
analyzed the obtained results using various visualization techniques. The analysis of
the processed genes expression values distributions allows concluding about high
effectiveness of the proposed technique, since its implementation allows allocating a
set of similarly distributed highly-expressed genes for the following processing.
References
1. Cha, Y.J., Park, S.M., You, R., Kim, H., Yoon, D.K.: Microstructure arrays of DNA using
topographic control. Nature Communications, 10(1), art. no. 2512 (2019) doi:
10.1038/s41467-019-10540-2
2. Nagashima, M., Miwa, N., Hirasawa, H., Katagiri, Y., Takamatsu, K., Morita, M.:
Genome-wide DNA methylation analysis in obese women predicts an epigenetic signature
for future endometrial cancer. Scientific Reports, 9(1) (2019), art. no. 6469 doi:
10.1038/s41598-019-42840-4
3. Depledge, D.P., Srinivas, K.P., Sadaoka, T., Bready, D., Mori, Y., Placantonakis, D.G.,
Mohr, I., Wilson, A.C.: Direct RNA sequencing on nanopore arrays redefines the
transcriptional complexity of a viral pathogen. Nature Communications, 10(1), art. no. 754
(2019) doi: 10.1038/s41467-019-08734-9
4. Lian, B., Hu, X., Shao, Z.-M.: Unveiling novel targets of paclitaxel resistance by single
molecule long-read RNA sequencing in breast cancer. Scientific Reports, 9(1), art. no. 6032
(2019) doi: 10.1038/s41598-019-42184-z
5. Bolstad, B.M., Irizarry, R.A., Åstrand, M., Speed, T.P.: A comparison of normalization
methods for high density oligonucleotide array data based on variance and bias.
Bioinformatics, 19 (2), pp. 185-193 (2003) doi: 10.1093/bioinformatics/19.2.185
6. Affymetrix. Statistical Algorithms Description Document. Affymetrix, Inc., Santa Clara,
CA, pp. 1-27 (2002)
7. Irizarry, R.A., Hobbs, B., Collin, F., Beazer-Barclay, Y.D., Antonellis, K.J., Scherf, U.,
Speed, T.P.: Exploration, normalization, and summaries of high-density oligonucleotide
array probe level data. Selected Works of Terry Speed, pp. 601-616 (2012) doi:
10.1007/978-1-4614-1347-9_15
8. Buermans, H.P.J., Ariyurek, Y., van Ommen, G., den Dunnen, J.T., 'T Hoen, P.A.C.: New
methods for next generation sequencing based microRNA expression profiling. BMC
Genomics, 11(1), art. no. 716 (2010) doi: 10.1186/1471-2164-11-716
9. Hackenberg, M., Sturm, M., Langenberger, D., Falcón-Pérez, J.M., Aransay, A.M.:
miRanalyzer: A microRNA detection and analysis tool for next-generation sequencing
experiments. Nucleic Acids Research, 37 (SUPPL. 2), pp. W68-W76 (2009) doi:
10.1093/nar/gkp347
10. Farazi, T.A., Brown, M., Morozov, P., ten Hoeve, J.J., Ben-Dov, I.Z., Hovestadt, V.,
Hafner, M., Renwick, N., Mihailović, A., Wessels, L.F.A., Tuschl, T.: Bioinformatic
analysis of barcoded cDNA libraries for small RNA profiling by next-generation
sequencing. Methods, 58 (2), pp. 171-187 (2012) doi: 10.1016/j.ymeth.2012.07.020
11. Hackenberg, M., Rodríguez-Ezpeleta, N., Aransay, A.M.: MiRanalyzer: An update on the
detection and analysis of microRNAs in high-throughput sequencing experiments. Nucleic
Acids Research, 39 (SUPPL. 2), pp. W132-W138 (2011) doi: 10.1093/nar/gkr247
12. Huang, P.-J., Liu, Y.-C., Lee, C.-C., Lin, W.-C., Gan, R.R.-C., Lyu, P.-C., Tang, P.:
DSAP: Deep-sequencing small RNA analysis pipeline. Nucleic Acids Research, 38
(SUPPL. 2), art. no. gkq392, pp. W385-W391 (2010) doi: 10.1093/nar/gkq392
13. Morin, R.D., O'Connor, M.D., Griffith, M., Kuchenbauer, F., Delaney, A., Prabhu, A.-L.,
Zhao, Y., McDonald, H., Zeng, T., Hirst, M., Eaves, C.J., Marra, M.A.: Application of
massively parallel sequencing to microRNA profiling and discovery in human embryonic
stem cells. Genome Research, 18 (4), pp. 610-621 (2008) doi: 10.1101/gr.7179508
14. Pantano, L., Estivill, X., Martí, E.: SeqBuster, a bioinformatic tool for the processing and
analysis of small RNAs datasets, reveals ubiquitous miRNA modifications in human
embryonic cells. Nucleic Acids Research, 38 (5), art. no. gkp1127, pp. e34.1-e34.13 (2009)
doi: 10.1093/nar/gkp1127
15. Li, Y., Zhang, Z., Liu, F., Vongsangnak, W., Jing, Q., Shen, B.: Performance comparison
and evaluation of software tools for microRNA deep-sequencing data analysis. Nucleic
Acids Research, 40 (10), pp. 4298-4305 (2012) doi: 10.1093/nar/gks043
16. Ronen, R., Gan, I., Modai, S., Sukacheov, A., Dror, G., Halperin, E., Shomron, N.:
miRNAkey: A software for microRNA deep sequencing analysis. Bioinformatics, 26 (20),
art. no. btq493, pp. 2615-2616 (2010) doi: 10.1093/bioinformatics/btq493
17. Dillies, M.-A., Rau, A., Aubert, J., at.al.: A comprehensive evaluation of normalization
methods for Illumina high-throughput RNA sequencing data analysis. Briefings in
Bioinformatics, 14 (6), art. no. bbs046, pp. 671-683 (2013) doi: 10.1093/bib/bbs046
18. Garmire, L.X., Subramaniam, S.: Evaluation of normalization methods in mammalian
microRNA-Seq data. RNA, 18 (6), pp. 1279-1288 (2012) doi: 10.1261/rna.030916.111
19. Izonin, I., Trostianchyn, A., Duriagina, Z., Tkachenko, R., Tepla, T., Lotoshynska, N.:
The combined use of the wiener polynomial and SVM for material classification task in
medical implants production. International Journal of Intelligent Systems and Applications,
10 (9), pp. 40-47 (2018) doi: 10.5815/ijisa.2018.09.05
20. Tam, S., Tsao, M.-S., McPherson, J.D.: Optimization of miRNA-seq data preprocessing.
Briefings in Bioinformatics, 16 (6), pp. 950-963 (2015) doi: 10.1093/bib/bbv019
21. Burov, Y., Vysotska, V., Kravets, P.: Ontological approach to plot analysis and modeling.
CEUR Workshop Proceedings, 2362, (2019)
22. Lytvyn, V., Vysotska, V., Dosyn, D., Burov, Y.: Method for ontology content and
structure optimization, provided by a weighted conceptual graph. Webology, 15 (2), pp. 66-
85 (2018)
23. Hu, Z., Mashtalir, S.V., Tyshchenko, O.K., Stolbovyi, M.I.: Clustering matrix sequences
based on the iterative dynamic time deformation procedure. International Journal of
Intelligent Systems and Applications, 10 (7), pp. 66-73 (2018) doi:
10.5815/ijisa.2018.07.07
24. Hu, Z., Bodyanskiy, Y.V., Tyshchenko, O.K., Tkachov, V.M.: Fuzzy clustering data
arrays with omitted observations. International Journal of Intelligent Systems and
Applications, 9 (6), pp. 24-32 (2017) doi: 10.5815/ijisa.2017.06.03
25. Babichev, S., Škvor, J., Fišer, J., Lytvynenko, V.: Technology of gene expression profiles
filtering based on wavelet analysis. International Journal of Intelligent Systems and
Applications, 10 (4), pp. 1-7 (2018) doi: 10.5815/ijisa.2018.04.01
26. Babichev, S.A., Gozhyj, A., Kornelyuk, A.I., Lytvynenko, V.I.: Objective clustering
inductive technology of gene expression profiles based on SOTA clustering algorithm.
Biopolymers and Cell, 33 (5), pp. 379-392 (2017) doi: 10.7124/bc.000961
27. NCBI Homepage, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE129336
28. Fernandes, J., Chandler, J.D., Lili, L.N., Uppal, K., Hu, X., Hao, L., Go, Y.-M., Jones,
D.P.: Transcriptome analysis reveals distinct responses to physiologic versus toxic
manganese exposure in human neuroblastoma cells. Frontiers in Genetics, 10 (JUN), art.
no. 676 (2019) doi: 10.3389/fgene.2019.00676
29. Robinson, M.D., Oshlack, A.: A scaling normalization method for differential expression
analysis of RNA-seq data. Genome Biology, 11(3), art. no. r25 (2010) doi: 10.1186/gb-
2010-11-3-r25
30. Babichev, S., Krejci, J., Bicanek, J., Lytvynenko, V.: Gene expression sequences
clustering based on the internal and external clustering quality criteria. Proceedings of the
12th International Scientific and Technical Conference on Computer Sciences and
Information Technologies, CSIT 2017, 1, art. no. 8098744, pp. 91-94 (2017) doi:
10.1109/STC-CSIT.2017.8098744
31. Babichev, S.: An evaluation of the information technology of gene expression profiles
processing stability for different levels of noise components. Data, 3 (4), art. no. 48 (2018)
doi: 10.3390/data3040048