geneEX

A Novel Tool to Assess Differential Expression

from Gene and Exon Sequencing Data

Orazio M Scicolone

∗

, Giulia Paciello

∗

and Elisa Ficarra

Department of Control and Computer Engineering, Politecnico di Torino, Torino, Italy

Keywords:

RNA-Sequencing, Differentially Expressed Genes, Exon Analysis, Table of Counts.

Abstract:

The widespread of Next Generation Sequencing technologies accounted in recent years for the possibility to

evaluate gene expression with great accuracy. Moreover, it allowed assessing differential gene expression

among biological conditions with high sensitivity. However, state-of-the-art bioinformatics methodologies for

differential gene expression evaluation from RNA Sequencing data still suffer from several drawbacks such

as reduced speciﬁcity. In this paper we propose geneEx, a novel methodology and tool for differential gene

expression evaluation from RNA Sequencing reads. By combining gene and exon expression evaluation and

BioMart information, geneEX provides users with annotated lists of highly reliable differentially expressed

genes. The results obtained in Sequencing Quality Control dataset proven the importance of a novel approach

to lower False Positive predictions from current methodologies and the strength of the proposed methodolog-

ical approach to increase the sensitivity of differentially expressed gene identiﬁcation.

1 INTRODUCTION

The advent of Next Generation Sequencing (NGS)

technologies dramatically reshaped genomics and

cancer genomics, allowing to produce huge amounts

of sequencing data with reduced per-base costs. Older

platforms such as microarrays were largely replaced

in the last decades by NGS techniques (Lee et al.,

2013). These techniques were exploited in several

contexts and with different purposes, such as the char-

acterisation of novel genomes, the deepening of par-

tially known genomic structures, or the identiﬁcation

of new variants at the base pair resolution.

Moreover, massive cDNA sequencing, also known

as RNA-Sequencing (RNA-Seq), was adopted to

carefully analyse and quantify transcriptomes, al-

lowing to discriminate differentially expressed genes

(DEGs) among biological conditions. To this aim,

several algorithms working on NGS data were de-

veloped. These algorithms implement different ap-

proaches for data normalization and DEG identiﬁca-

tion, as widely discussed in (Young et al., 2012). Six

of these methods, i.e. Cuffdiff (Trapnell et al., 2013),

edgeR (Robinson et al., 2010), DESeq (Love et al.,

2014), PoissonSeq (Li et al., 2012), baySeq (Hardcas-

tle and Kelly, 2010) and limma (Smyth, 2005), were

∗

co-ﬁrst-author

recently compared in SEQC (DeLuca et al., 2012) and

ENCODE (Consortium et al., 2004) datasets consid-

ering, as examples, their normalisation and speciﬁcity

performance. Overall, no speciﬁc method was proven

to be the best solution in all the comparisons (Rapa-

port et al., 2013).

To the light of this consideration, as stated in (Ra-

jkumar et al., 2015), the combined adoption of dif-

ferent algorithms for differential gene expression as-

sessment is highly advisable to achieve good sensi-

tivity. However, it has to be considered that higher

the number of tools adopted higher both the compu-

tational costs and the amount of False Positive (FP)

predictions obtained in output.

To overcome limitations proper of state-of-the-art

methodologies for DEG identiﬁcation, we propose

a novel methodology and tool named geneEX. ge-

neEX comes in the form of an R package that can

be easily integrated within all of the bioinformatics

pipelines working on RNA-Seq data. It performs

DEG assessment by processing and integrating the

results from three widely adopted R packages, i.e.,

DESeq2, edgeR, and DEXSeq. Speciﬁcally, the ﬁrst

two methods are exploited by geneEX to perform

expression evaluation at the gene level, whereas the

last one to make exon expression assessment. ge-

neEX implements a series of elaboration and ﬁltering

174

Scicolone, O., Paciello, G. and Ficarra, E.

geneEx - A Novel Tool to Assess Differential Expression from Gene and Exon Sequencing Data.

DOI: 10.5220/0006657501740181

In Proceedings of the 11th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2018) - Volume 3: BIOINFORMATICS, pages 174-181

ISBN: 978-989-758-280-6

stages which allow to shrink down the list of DEGs

from current methodologies, focusing on the more re-

liable ones. By integrating annotations from BioMart

database (Durinck et al., 2005), geneEX provides bi-

ologists with a series of information useful to fur-

ther prioritise true DEGs. Moreover, its analysis can

be easily triggered according to both user needs and

computational resources. The novel geneEX will be

released as soon as possible as a Bioconductor pack-

age (Huber et al., 2015).

We assessed the need for a novel DEG analy-

sis tool and geneEX performance in the Sequencing

Quality Control (SEQC) dataset (Shi et al., ; Shi et al.,

2006). Results from these analyses proven the impor-

tance of the ﬁltering stages implemented within ge-

neEX to lower FPs, thus increasing the speciﬁcity of

the detection and providing users with highly reliable

DEG candidates.

2 MATERIALS AND METHODS

2.1 Dataset

Data exploited in our study was downloaded from

SRA database with accession codes SRX333347-

SRX333356. The considered dataset is part of SEQC

study and comprises two groups of samples, which

will be referred to as Group 1 and Group 2 in the fol-

lowing. Speciﬁcally, the ﬁve samples belonging to

the ﬁrst group are replicates of the Stratagene Uni-

versal Human Reference RNA (UHRR) that contains

the RNA from ten human cell lines with 2% by vol-

ume of synthetic RNAs from the External RNA Con-

trol Consortium (ERCC) mix1. Conversely, Group

2 comprises ﬁve replicates of the Ambion’s Human

Brain Reference RNA (HBRR) with 2% by volume

of ERCC mix2. ERCC spike-in control mixes are sets

of 92 250-2000 nt (nucleotide) long polyadenylated

transcripts from the ERCC plasmid reference library.

These spike-in sequences can be further grouped into

four subgroups, with well deﬁned molar concentra-

tions in mix1 and mix2.

2.2 Read Alignment and ene/xon Count

Calculation

According to (Rapaport et al., 2013) study, paired-end

reads were aligned using tophat2 (Kim et al., 2013)

on a reference sequence comprising both human

chromosomes and ERCC spike-in nucleotide series.

UCSC gene and ERCC spike-in counts were retrieved

from tophat2 output ﬁles by using HTSeq (Anders

et al., 2015). Tophat2 and HTSeq parameters were

tuned as stated in (Rapaport et al., 2013). Exon counts

were obtained by running DEXSeq dexseq count.py

program with default parameters. The gtf annotation

ﬁle exploited during tophat2, HTSeq and DEXseq

analyses comprises both spike-in and UCSC gene se-

quences, according to (Rapaport et al., 2013).

2.3 geneEX Workﬂow

geneEX tool is an automatic, parallel and highly cus-

tomizable R module for differential gene expression

analysis that works on RNA-Seq data.

It is built on-top of three widely adopted state-of-

the-art Bioconductor (Gentleman et al., 2004) pack-

ages for differential gene expression evaluation, i.e.,

DESeq2 (Love et al., 2014), edgeR (Robinson et al.,

2010) and DEXSeq (Li et al., 2015). These packages

provide a method to test for differential gene/exon ex-

pression among biological conditions by exploiting a

statistical model to describe data distribution. Specif-

ically, DESeq2 and DEXSeq model data using a neg-

ative binomial distribution, whereas edgeR ﬁts data

using the Poisson distribution.

The input data of these algorithms is represented

by the so called tables of counts. The table of

counts reports, for each gene or exon in the genome,

the number of reads (i.e., the elementary sequences

output of Next Generation Sequencing machines)

mapped on it. In more detail, edgeR was designed

to identify differentially expressed genes or exons,

DESeq2 to retrieve differentially expressed genes and

DEXSeq to identify exon usage biases among condi-

tions.

geneEX elaborates and combines the results from

these three tools to provide users with a reduced list

of highly reliable DEGs. Moreover, geneEX anno-

tates these genes with a series of information from

BioMart (Smedley et al., 2015) database, facilitating

they further prioritization by biologist and clinician

investigation.

Speciﬁcally, geneEX makes use of DESeq2 and

edgeR to test for differential genes, whereas of

DEXSeq to further prioritized these candidates based

on exon expression data.

geneEX workﬂow is depicted in Figure 1 and

detailed in the following. Rectangular white boxes

identify those activities implemented by state-of-the-

art bioinformatics algorithms, whereas the grey ones

those performed by ad-hoc designed R scripts. Con-

versely, yellow, pink and orange irregular shapes re-

port respectively on the input data, the output ﬁles

from the differential analysis and the results from

Rlog normalisation.

geneEx - A Novel Tool to Assess Differential Expression from Gene and Exon Sequencing Data

175

Figure 1: geneEX ﬂowchart.

Starting point of geneEX analysis is the setting of

the tool parameters within a conﬁguration ﬁle (block

A in Figure 1).

Users have to specify i) the path of the folder

that stores the input table of counts organized ac-

cording to the biological condition, ii) the path of

an annotation ﬁle that is exploited by DEXSeq to

create a DEXSeq object, iii) the number of threads

to be launched during the analysis, iv) the statisti-

cal measures, i.e., p-value (pval), adjusted p-value

(adj-pval), False Discovery Rate (FDR) or logarith-

mic Fold Change (log

FC) and their values to be

adopted by DESeq2, edgeR and DEXSeq to ﬁlter out

data, and iv) the number of differentially expressed

exons from DEXSeq results to be considered to fur-

ther ﬁlter out data. As shown in block C of Figure

1, the input gene/exon table of counts are processed

by geneEX to generate DESeq2, edgeR and DEXSeq

objects. These objects represent the input required

by DESeq2, edgeR and DEXSeq to perform differen-

tial gene/exon expression analysis. Additionally, gene

counts are normalized by DESeq2 RLog Transform

function (block B in Figure 1) which converts the raw

counts into log2-scale data. This to minimise the dif-

ferences among samples for low expressed genes and

to normalise data based on library sizes. RLog trans-

formation represents a fundamental preliminary step

for different analysis techniques such as clustering or

PCA (Love et al., 2014). These methods can be eas-

ily integrated within the novel implemented package,

making geneEX a very versatile tool.

The next phase of geneEX, shown in block D of

Figure 1, consists in the execution of DESeq2, edgeR

and DEXSeq algorithms. The output of both DESeq2

and edgeR runs is a list of genes, where each gene

comes with several statistical scores such as pval,

adj-pval, FDR and log

FC, that describe the conﬁ-

dence in assessing differential gene expression. Sim-

ilarly, these results are provided in terms of exons by

DEXSeq. These three output ﬁles (i.e., DESeq2 re-

sults, DEXSeq results and edgeR results in Figure 1)

are saved in CSV ﬁles.

Step E implements the ﬁrst ﬁltering activity to be

performed on the output lists from DESeq2, edgeR

BIOINFORMATICS 2018 - 9th International Conference on Bioinformatics Models, Methods and Algorithms

176

and DEXSeq tools. Speciﬁcally, results are ﬁltered

according to the ﬁxed thresholds and saved into CSV

ﬁles. In F, the gene lists from DESeq2 and edgeR

analyses are further elaborated, by retaining those

genes with at least a given number of exons (ﬁxed in

the parameter setting activity) that are with high prob-

ability differentially expressed (according to DEXSeq

analysis and the adopted ﬁltering threshold). Even in

this case results are provided to users in CSV format.

It is worth noting that users are let free to select the

statistical measure (and its value) to ﬁlter data based

on exon expression. This allows to implement a ﬁl-

tering step that considers the speciﬁc protocol adopted

for RNA extraction (in most of cases poly(A) selection

or rRNA depletion). Finally, in step G, those genes

shared by the last two ﬁles are extracted and saved in

a CSV ﬁle. The whole lists of genes contained in both

the intermediate and ﬁnal output ﬁles are annotated

in the different processing steps by querying Biomart

database(Smedley et al., 2015). Thus, each candidate

gene will be provided to the user with a series of infor-

mation such as the chromosome to which it belongs

and the relative chromosome band, its start and end

positions on the chromosome and the HUGO Gene

Nomenclature Committee (HGNC) symbol. All this

information is essential to further prioritise true dif-

ferentially expressed genes based on the expertise of

biologists and clinicians.

3 RESULTS AND DISCUSSION

We used ERCC spike-in synthetic sequences to assess

the performance of DESeq2 and edgeR tools, on-top

of which geneEX is built. These results allowed to

point out the need for ad-hoc ﬁltering strategies to fo-

cus on a reduced list of highly reliable DEGs to be fur-

ther deepened by wet-lab experiments. Conversely,

non-synthetic reads from SEQC (UHRR and HBRR

datasets) were exploited to test the whole geneEX al-

gorithm and to discuss the results it provided.

3.1 ERCC Spike-in Analysis

ERCC spike-in synthetic sequences cannot be di-

vided into exons. Thus, reads deriving from their se-

quencing are not suitable for geneEX analysis, which

comprises DEXSeq processing. However, ERCC

dataset can be used as a benchmark to evaluate both

edgeR and DESeq2 performance. We ran DESeq2

and edgeR with default parameters comparing Group

1 and Group 2. Speciﬁcally, we focused on those

spike-in sequences not differentially expressed since

characterised by equal molar concentration in both

mix1 and mix2 (ERCC subgroup ID: B). The adj-

pvals provided by DESeq2 and the FDRs from edgeR

for these 23 not differentially expressed ERCC se-

quences were used to build the respective Receiver

Operating Characteristic (ROC) curves and to calcu-

late the relative Area Under Curves (AUCs). The

blue and yellow ROC curves of Figure 3 report on

DESeq2 and edgeR results respectively. Both tools

ensured sensitivity levels close to 0.8 with a speci-

ﬁcity of about 0.56 when thresholds equal to 0.034

and 0.01 are applied to the adj-pvals from DESeq2

and the FDRs from edgeR. These results conﬁrmed

the importance of ad-hoc ﬁltering strategies based

on statistical measures such as the pval, to obtain

satisfactory sensitivity and speciﬁcity levels. More-

over, we assessed DESeq2 and edgeR capability to

correctly identify ERCC differentially expressed se-

quences (ERCC subgroup IDs: A, C and D) when ap-

plying different adj-pval and FDR thresholds. These

sequences were selected according to a log

FC6= 0

between Group 1 and Group 2. 69 out of 92 ERCC

sequences were identiﬁed as differentially expressed.

Two analyses were performed, using as input for DE-

Seq2 and edgeR run a table of counts containing i)

ERCC sequences only or ii) the whole list of hg19

annotated genes and the complete set of ERCC se-

quences. This twofold analysis allowed us to discuss

both the obtained results in terms of False Positive

(FP), False Negative (FN), True Negative (TN) and

True Positive (TP) predictions and the strength of the

statistical model applied by DESeq2 and edgeR. Fig-

ure 2 reports on these results. Speciﬁcally, each Venn

diagram depicts in the circle labelled as ERCC the

number of differentially expressed sequences deﬁned

in ERCC experiment, whereas in the circles named

DESeq2 and edgeR the amount of differentially ex-

pressed sequences identiﬁed by DESeq2 and edgeR

from RNA-Seq data. The different subﬁgures in the

ﬁrst row of Figure 2 are relative to the former analy-

sis, whereas the second row reports on the latter anal-

ysis.

With reference to the ﬁrst analysis (that involved

ERCC sequences only), this allowed to carefully eval-

uate the performance of DESeq2 and edgeR in terms

of sensitivity and speciﬁcity. With the sensitivity cal-

culated as the ratio between the amount of TPs and the

sum of TP and FN predictions, whereas the speciﬁcity

as the ratio between TNs and the sum of TN and FP

predictions. This analysis allowed also to assess the

convenience of a DESeq2-edgeR combined approach

for DEG identiﬁcation. Speciﬁcally, sequences iden-

tiﬁed by one or both tools and shared with ERCC set

are TPs, whereas those not shared with ERCC set are

FP predictions. Sequences included in ERCC dataset

geneEx - A Novel Tool to Assess Differential Expression from Gene and Exon Sequencing Data

177

Figure 2: Consensus among in-silico and wet-lab methodologies on ERCC spike-in differentially expressed sequence identi-

ﬁcation. Subﬁgures A, B and C report on the results obtained when considering ERCC sequences only, whereas Subﬁgures

D, E and F depict those provided by the adopted methodologies on the whole list of hg19 annotated genes and the complete

set of ERCC sequences.

only are FN predictions, and those not contained in

ERCC set (non differentially expressed) and not iden-

tiﬁed by the tools are TNs. These data are shown in

Table 1.

The different rows of Table 1 report on DESeq2,

edgeR and DESeq2-edgeR results respectively. The

adoption of DESeq2 ensured the best results in terms

of sensitivity and speciﬁcity. However, it has to be

considered that the number of sequences involved in

this analysis is very reduced, far from the number of

genes considered in standard experiments. As conse-

quence, the reduced number of FP predictions is not

meaningful, making useless an approach that com-

bines DESeq2 and edgeR tools since negatively im-

pacting on the sensitivity while not improving speci-

ﬁcity. However, as clearly stated in (Rapaport et al.,

2013), in real cases the number of FPs is gener-

ally very huge, calling for ad-hoc ﬁltering procedures

to shrink down their numbers. This aspect clearly

emerged in the second analysis, whose results are re-

ported in the second row of Figure 2. The lack of

a complete wet-lab validation for all genes involved

in the experiment makes impossible the calculation

of the overall number of FP, TP, FN and TN predic-

tions. However, the number of TP ERCC sequences

identiﬁed by DESeq2 and edgeR is higher than that

reported as result of the ﬁrst analysis. Conﬁrming the

importance of a statistically meaningful input to build

the probability distribution exploited by DESeq2 and

edgeR tools. Moreover, with high likelihood, not all

the genes labelled as differential by DESeq2, edgeR

or DESeq2-edgeR are in reality TPs.

This conﬁrms the need for an accurate selection of

the statistical measure and its value to be adopted to

lower FP rates while maintaining satisfactory sensi-

tivity levels, keeping in mind that too strict thresholds

can negatively impact on sensitivity while poorly im-

proving speciﬁcity. By exploiting exon analysis, ge-

neEX accounts for the possibility to adopt more re-

laxed thresholds that preserve sensitivity while lower-

ing the number of FPs.

3.2 SEQC Non Synthetic Data Analysis

In (Canales et al., 2006), three different quantitative

gene expression measurement technologies were as-

sessed, i.e. TaqMan Gene Expression Assay, Stan-

dardized RT (Sta)RT-PCR assay and QuantiGene as-

say, and the obtained results compared with those

from DNA microarray platforms. Speciﬁcally, Taq-

Man assays were performed on 997 genes. We ex-

ploited this set of genes to evaluate geneEX per-

formance, deﬁning as DEGs those genes having a

log

FC > 0.5 or < −0.5 between Group 1 and Group

2 according to qRT-PCR analysis. In more detail, our

analysis comprised four replicates from the UHRR

for Group 1, whereas four replicates of the HBRR

for Group 2. 30 out 997 genes were excluded from

our experimental setup because not found in BioMart

database. Based on the adopted log

FC, 742 genes

were marked as differentially expressed, whereas 225

as a not differential.

We ran DESeq2, edgeR and geneEX on the 967

genes from TaqMan assays imposing default param-

BIOINFORMATICS 2018 - 9th International Conference on Bioinformatics Models, Methods and Algorithms

178

Table 1: Sensitivity/Speciﬁcity levels of DESeq2, edgeR and DESeq2-edgeR .

FDR

edgeR

/adj-pval

DESeq2

=0.1 FDR

edgeR

/adj-pval

DESeq2

=0.05 FDR

edgeR

/adj-pval

DESeq2

=0.01

Sensitivity Speciﬁcity Sensitivity Speciﬁcity T Sensitivity Speciﬁcity

DESeq2 73.9% 47.8% 72.4% 60.9% 63.8% 73.9%

edgeR 73.9% 34.8% 66.7% 34.8% 53.6% 47.8%

edgeR ∩ DESeq2 66.6% 47.8% 63.8% 60.9% 53.6% 73.9%

Figure 3: ROC curves of DESeq2 and edgeR tools.

eters. The obtained results were analysed to assess

both the speciﬁcity/sensitivity of these methods when

used singly and those of a DESeq2-edgeR combined

approach. Speciﬁcally, two different statistical mea-

sures were adopted to identify the DEGs, i.e. the pval

and the log

FC. In the ﬁrst analysis, based on pval

scores, we used as ﬁltering threshold for DESeq2,

edgeR and geneEX, a value of 0.05. With reference

to geneEX, this threshold was applied to the all the

ﬁltering stages implemented by the tools on-top of

which it is built (i.e., Block E in Figure 1). Simi-

larly, in the second analysis, a log

FC of 0.5 was im-

posed to focus on highly reliable DEGs. Figure 4 re-

ports on the number of DEGs shared among the com-

pared in-silico methodologies and the TaqMan assay

experimental results. Speciﬁcally, Subﬁgures 4.A and

4.C show respectively the amount of DEGs identi-

ﬁed by TaqMan assay, DESeq2 and edgeR (Subﬁg-

ure A) or by TaqMan assay and geneEX (Subﬁgure

C), when imposing the pval ﬁltering threshold. Sim-

ilarly, Subﬁgures 4.B and 4.D report on the num-

ber of genes identiﬁed by the same methodologies

when applying the log

FC threshold. With reference

to Subﬁgures 4.A and 4.B, the adoption of a differ-

ent statistical measure for data ﬁltering, strongly im-

pacted on the obtained results. Indeed, by ﬁltering us-

ing a pval threshold we obtained a higher percentage

Figure 4: Consensus among in-silico and wet-lab method-

ologies on UHRR and HBRR DEGs identiﬁcation. Subﬁg-

ures A and C report on the amount of DEGs provided by the

adopted approaches when using the pval as statistical mea-

sure to ﬁlter data. Similarly, Subﬁgures B and D depict the

results provided by the same methodologies when ﬁltering

according to a log

FC threshold.

of TP predictions, but at the same time a signiﬁcant

amount of FP ones. Moreover, it has to be noticed

that, in both cases, even DESeq2-edgeR combined ap-

proach resulted in conspicuous amounts of FPs. ge-

neEX tries to lower the number of FPs by evaluating

exon expression data from DEXSeq analysis. As ex-

pected, the introduction of an additional ﬁltering step,

negatively impacted on the number of identiﬁed true

DEGs. Indeed, when considering geneEX pval and

log

FC thresholds, we lost respectively about 54%

and 4.5% of TPs with respect to DESeq2-edgeR com-

bined approach results. However, geneEX analysis

also lowered the number of FPs, as desired. Account-

ing for a decrease of about 11.5% and 1.2% when ﬁl-

tering using the pval and log

FC thresholds respec-

tively. Results from the previous analyses were fur-

ther elaborated to compute the sensitivity and speci-

ﬁcity values associated with the adoption of the differ-

ent methodologies and ﬁltering approaches. Table 2

reports on this data. Speciﬁcally, the different rows of

Table 2 show respectively DESeq2, edgeR, DESeq2-

edgeR combined approach and geneEX results. As

geneEx - A Novel Tool to Assess Differential Expression from Gene and Exon Sequencing Data

179

Table 2: Sensitivity/Speciﬁcity levels of DESeq2, edgeR, DESeq2-edgeR and geneEX .

pvalue

DESeq2,edgeR

<0.05 log2FC

DESeq2,edgeR

<0.5 pvalue

DESeq2,edgeR,DEX Seq

<0.05 log2FC

DESeq2,edgeR,DEX Seq

<0.5

Sensitivity Speciﬁcity Sensitivity Speciﬁcity Sensitivity Speciﬁcity Sensitivity Speciﬁcity

DESeq2 93.5% 29.8% 90.2% 57.8% - - - -

edgeR 94.0% 31.6% 90.7% 60 % - - - -

edger ∩ DESeq2 93.1% 38.7% 89.9% 63.1% - - - -

geneEX - - - - 22.2% 88 % 84.0% 68.4%

previously discussed in terms of TPs, FPs, FNs and

TNs, the adoption of a single tool working at the gene-

expression level did not ensure satisfactory speci-

ﬁcity. With values ranging from 29.8% to 60% de-

pending on the adopted ﬁltering threshold (rows 1 and

2 in Table 2 ). The speciﬁcity slightly increased when

adopting a DESeq2-edgeR combined approach, still

maintaining a high sensitivity level. With reference

to geneEX (row 4 of Table 2), its adoption accounted

for an increase in the speciﬁcity levels compared to all

the other tested methods and independently from the

adopted ﬁltering measure. However, it is worth not-

ing that the higher improvement was achieved when

ﬁltering according to the pval (speciﬁcity equal to

88%). However, this strongly impacted on the sen-

sitivity, reduced to the 22%. A higher sensitivity level

(84%) was instead ensured when ﬁltering according

to the log

FC measure while preserving a speciﬁcity

level (68.4%) at least 5% higher than that provided by

the compared methodologies. These analyses proved

i) the importance of ﬁltering stages based on exon

analysis to improve speciﬁcity and ii) the importance

of an accurate choice of the statistical measure to be

adopted during the different ﬁltering steps. Moreover,

we would like to underline that, as already observed

in the context of ERCC sequence study, the number

of genes investigated in this analysis is very reduced,

due to the lack of datasets coming with a full experi-

mental validation. As consequence, even the number

of FP predictions from the different tools is not so

high. However, in normal experiments FPs are nu-

merous, making impossible the wet-lab validation of

all the results (that comprise both TPs and FPs) from

in-silico analysis and calling for ad-hoc procedures to

shrink the number of candidates focusing on the most

reliable ones.

4 CONCLUSIONS

In this paper, we propose a novel methodology and

tool, namely geneEX, for DEG identiﬁcation. ge-

neEX integrates the results provided by three widely

adopted bioinformatics algorithms for DEG analysis.

Speciﬁcally, it uses DESeq2 and edgeR tools to iden-

tify highly reliable DEG, whereas DEXSeq to further

prioritise these candidates based on exon expression

data. Moreover, by interrogating BioMart database,

geneEX annotates each candidate gene with a series

of information that can be exploited by biologists or

clinicians to focus on the most signiﬁcant predictions.

Thus, geneEX provides users with a reduced list of

statistically meaningful and highly reliable DEGs to

be deeply investigated by wet-lab experiments. ge-

neEX is highly customizable and easy to use. Users

can trigger its run by specifying both the statistical

measures and the relative values to be adopted dur-

ing DESeq2, edgeR and DEXSeq ﬁltering stages. We

assessed the need for a novel DEG analysis tool on

the ERCC sequences from SEQC dataset. Indeed, this

analysis allowed to highlight the limits of DEG analy-

sis tools, which output lists are generally plagued with

huge amounts of FP predictions, overwhelming wet-

lab validation possibilities. Conversely, geneEX per-

formance was evaluated on four replicates of UHRR

and HBRR datasets, proving the advantages in terms

of speciﬁcity increase associated with its adoption.

REFERENCES

Anders, S., Pyl, P. T., and Huber, W. (2015). Htseqa python

framework to work with high-throughput sequencing

data. Bioinformatics, 31(2):166–169.

Canales, R. D., Luo, Y., Willey, J. C., Austermiller, B., Bar-

bacioru, C. C., Boysen, C., Hunkapiller, K., Jensen,

R. V., Knight, C. R., Lee, K. Y., et al. (2006). Eval-

uation of dna microarray results with quantitative

gene expression platforms. Nature biotechnology,

24(9):1115.

Consortium, E. P. et al. (2004). The encode (encyclope-

dia of dna elements) project. Science, 306(5696):636–

640.

DeLuca, D. S., Levin, J. Z., Sivachenko, A., Fennell, T.,

Nazaire, M.-D., Williams, C., Reich, M., Winckler,

W., and Getz, G. (2012). Rna-seqc: Rna-seq metrics

for quality control and process optimization. Bioinfor-

matics, 28(11):1530–1532.

Durinck, S., Moreau, Y., Kasprzyk, A., Davis, S., De Moor,

B., Brazma, A., and Huber, W. (2005). Biomart

and bioconductor: a powerful link between biological

databases and microarray data analysis. Bioinformat-

ics, 21(16):3439–3440.

Gentleman, R. C., Carey, V. J., Bates, D. M., Bolstad, B.,

Dettling, M., Dudoit, S., Ellis, B., Gautier, L., Ge, Y.,

Gentry, J., et al. (2004). Bioconductor: open software

BIOINFORMATICS 2018 - 9th International Conference on Bioinformatics Models, Methods and Algorithms

180

development for computational biology and bioinfor-

matics. Genome biology, 5(10):R80.

Hardcastle, T. J. and Kelly, K. A. (2010). bayseq: empir-

ical bayesian methods for identifying differential ex-

pression in sequence count data. BMC bioinformatics,

11(1):422.

Huber, W., Carey, V. J., Gentleman, R., Anders, S., Carlson,

M., Carvalho, B. S., Bravo, H. C., Davis, S., Gatto, L.,

Girke, T., et al. (2015). Orchestrating high-throughput

genomic analysis with bioconductor. Nature methods,

12(2):115–121.

Kim, D., Pertea, G., Trapnell, C., Pimentel, H., Kelley,

R., and Salzberg, S. L. (2013). Tophat2: accurate

alignment of transcriptomes in the presence of inser-

tions, deletions and gene fusions. Genome biology,

14(4):R36.

Lee, C.-Y., Chiu, Y.-C., Wang, L.-B., Kuo, Y.-L., Chuang,

E. Y., Lai, L.-C., and Tsai, M.-H. (2013). Common ap-

plications of next-generation sequencing technologies

in genomic research. Translational Cancer Research,

2(1):33–45.

Li, J., Witten, D. M., Johnstone, I. M., and Tibshirani, R.

(2012). Normalization, testing, and false discovery

rate estimation for rna-sequencing data. Biostatistics,

13(3):523–538.

Li, Y., Rao, X., Mattox, W. W., Amos, C. I., and Liu, B.

(2015). Rna-seq analysis of differential splice junc-

tion usage and intron retentions by dexseq. PloS one,

10(9):e0136653.

Love, M. I., Huber, W., and Anders, S. (2014). Moderated

estimation of fold change and dispersion for rna-seq

data with deseq2. Genome biology, 15(12):550.

Rajkumar, A. P., Qvist, P., Lazarus, R., Lescai, F., Ju, J.,

Nyegaard, M., Mors, O., Børglum, A. D., Li, Q., and

Christensen, J. H. (2015). Experimental validation of

methods for differential gene expression analysis and

sample pooling in rna-seq. BMC genomics, 16(1):548.

Rapaport, F., Khanin, R., Liang, Y., Pirun, M., Krek, A.,

Zumbo, P., Mason, C. E., Socci, N. D., and Betel, D.

(2013). Comprehensive evaluation of differential gene

expression analysis methods for rna-seq data. Genome

biology, 14(9):3158.

Robinson, M. D., McCarthy, D. J., and Smyth, G. K. (2010).

edger: a bioconductor package for differential expres-

sion analysis of digital gene expression data. Bioin-

formatics, 26(1):139–140.

Shi, L., Campbell, G., Jones, W., Campagne, F., Wen, Z.,

Walker, S., Su, Z., and Chu, T. Goodsaid, 373 fm,

pusztai, l., et al.(2010). the microarray quality control

(maqc)-ii study of common practices 374 for the de-

velopment and validation of microarray-based predic-

tive models. Nature Biotechnology, 375:28.

Shi, L., Reid, L., Jones, W., Shippy, R., Warrington, J.,

Baker, S., Collins, P., and de Longueville, F. (2006).

Es et al. kawasaki, and maqc consortium. the microar-

ray quality control (maqc) project shows inter-and in-

traplatform reproducibility of gene expression mea-

surements. Nat Biotechnol, 24:1151–61.

Smedley, D., Haider, S., Durinck, S., Pandini, L., Provero,

P., Allen, J., Arnaiz, O., Awedh, M. H., Baldock,

R., Barbiera, G., et al. (2015). The biomart com-

munity portal: an innovative alternative to large, cen-

tralized data repositories. Nucleic acids research,

43(W1):W589–W598.

Smyth, G. K. (2005). Limma: linear models for microar-

ray data. In Bioinformatics and computational biology

solutions using R and Bioconductor, pages 397–420.

Springer.

Trapnell, C., Hendrickson, D. G., Sauvageau, M., Goff, L.,

Rinn, J. L., and Pachter, L. (2013). Differential analy-

sis of gene regulation at transcript resolution with rna-

seq. Nature biotechnology, 31(1):46–53.

Young, M. D., McCarthy, D. J., Wakeﬁeld, M. J., Smyth,

G. K., Oshlack, A., and Robinson, M. D. (2012).

Differential expression for rna sequencing (rna-seq)

data: mapping, summarization, statistical analysis,

and experimental design. In Bioinformatics for High

Throughput Sequencing, pages 169–190. Springer.

geneEx - A Novel Tool to Assess Differential Expression from Gene and Exon Sequencing Data

181