Replicability of Differentially Expressed Genes Versus Biological

Pathways Biomarkers in Diagnosing Sepsis

Kelsey Winkeler

1 a

and Carly A. Bobak

2 b

Department of Psychology, Virginia Polytechnic Institute and State University Blacksburg, VA 24060, U.S.A.

Biomedical Data Science, Dartmouth College, Hanover, NH 03755, U.S.A.

Keywords:

Differentially Expressed Genes, Sepsis, Replicability, Biological Pathways.

Abstract:

It is generally believed that biological pathways representing curated gene sets are not only more interpretable,

but also more replicable and reproducible than gene signatures. With the falling costs of next generation se-

quencing, we are approaching a point where the cost fully sequencing the transcriptome is competitive with

quantifying a targeted gene expression signature which opens up the possibility of pathway signatures for in-

fectious disease. In this work, we evaluated if pathway based signatures are really more reproducible than gene

signatures (improvement between 0.83 and over 1 million fold), and amend a meta-analysis framework known

for generating highly reproducible gene signatures to instead produce pathway signatures (AUC improves

from 0.854 to 0.964 and 0.556 to 0.677 between gene and pathway signatures in independent validation data).

We conclude that pathway based signatures show clinical promise for the diagnosis of infectious disease, and

there is a growing need for methods considering such signatures.

1 INTRODUCTION

Reproducibility and replicability, wherein discrimi-

natory biological features are consistently associated

with a phenotype both within the same dataset and

across new datasets, is a major challenge to using

differentially expressed genes (DEGs) for diagnoses

(Crow et al., 2019; Sweeney et al., 2015). Most ap-

proaches take minimal biological information into ac-

count, struggle to remain consistent across samples

and platforms, and make data difﬁcult to interpret bi-

ologically. (Tan, 2003; Zhang et al., 2008) Recent

efforts have been made to combine gene expression

with knowledge of biological pathways and function

using Gene Ontology (GO) terms and other annotated

pathways. These methods have the advantage of be-

ing less complex and more biologically interpretable

than traditional analysis of DEGs, emphasizing net-

works of related genes over individual genes. (Khatri

et al., 2012; Zhang et al., 2009)

There has been some successes: Zarringhalam et

al. predicted kidney transplant rejection and response

to Inﬂiximab in ulcerative colitis (Zarringhalam et al.,

2014), and Pradines et al. found improved repro-

https://orcid.org/0000-0002-9452-9428

https://orcid.org/0000-0001-8631-4753

ducibility within and between datasets in several dis-

eases (Pradines et al., 2020). Such approached may

be useful even in studies with small sample sizes.

(Lim et al., 2015) While prior attempts at diagnostics

aim to reduce biomarker quantity to make any result-

ing assay more cost-effective to produce, the falling

cost of transcriptome-wide sequencing makes these

proposed pathway signatures possible.(Alpern et al.,

2019; Mayday et al., 2019; Sholder et al., 2020)

Sepsis is a particularly important potential appli-

cation, as it can present similarly to the non-infectious

systemic inﬂammatory response syndrome (SIRS)

and lacks a rapid, gold-standard diagnostic. Begin-

ning treatment within the ﬁrst ‘golden hour’ is integral

for reducing mortality in severe sepsis; however, inap-

propriate and overuse of antibiotics in hospitals con-

tinues to increase rates of MRSA and other antibiotic-

resistant microbes. (Ferrer et al., 2014; van Zanten,

2014; Cohen et al., 2015; Sweeney and Khatri, 2017)

Attempts to create a sepsis diagnostic DEG signa-

ture include the sepsis meta score (Sweeney et al.,

2015), FA1M3:PLAC8 ratio (Scicluna et al., 2015),

and the SeptiCyte lab (McHugh et al., 2015), the three

of which were compared using 39 public data sets

in Sweeney and Khatri in 2017(Sweeney and Khatri,

2017) and demonstrated a similar ability to discrimi-

nate between patients with and without sepsis.

160

Winkeler, K. and Bobak, C.

Replicability of Differentially Expressed Genes Versus Biological Pathways Biomarkers in Diagnosing Sepsis.

DOI: 10.5220/0010976300003123

In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 3: BIOINFORMATICS, pages 160-167

ISBN: 978-989-758-552-4; ISSN: 2184-4305

In this work, we extend the current research com-

paring pathway biomarkers to DEGs by speciﬁcally

focusing on the implications of pathway biomarkers

in diagnosis pipelines. Speciﬁcally, we consider com-

paring the overall rank of important DEGs to the rank

of important pathways across multiple datasets col-

lected on multiple gene expression array platforms,

and evaluate how a pathway based approach could be

amended to a meta-analysis framework to further im-

prove replicability of diagnostic biomarkers.

2 METHODS

All analyses were performed in R version 4.0.3. (R

Core Team, 2017)

2.1 Data Sources

Following Sweeney et al.(Sweeney and Khatri, 2017),

we downloaded publicly available human gene ex-

pression microarray datasets from the NIH Gene

Expression Omnibus (GEO). (Barrett et al., 2012;

Sutherland et al., 2011; Dolinay et al., 2012; Par-

nell et al., 2012; McHugh et al., 2015; Kim et al.,

2021) Cases included blood samples from adult pa-

tients with severe infection or sepsis, taken on day

1 of presentation and pre-treatment. Controls were

limited to hospitalization patients without infectious

diagnoses. A summary of the included datasets is

shown in Table 1. Of the discovered datasets, two

were withheld for diagnostic validation while three

were used for discovery purposes.

2.2 Processing and Biomarker Analysis

Normalized, publicly available datasets were

downloaded using the ‘MetaIntegrator’ package in

R.(Haynes et al., 2016) Probe IDs were matched to

HUGO gene symbol; where multiple probes matched

to the same symbol, the median expression value was

used. (Tweedie et al., 2021) Differentially expressed

genes (DEGs) were identiﬁed using the ‘limma’

package in R. (Ritchie et al., 2015)

We conducted pathway analysis using two meth-

ods. First, we conducted Gene Set Enrichment Anal-

ysis (GSEA) using the results from the DEGs de-

scribed above. (Mootha et al., 2003; Subrama-

nian et al., 2005) Genes were ﬁrst ranked using

sign( f oldchange) × (− log

(p-value)) as described

previously (Chen et al., 2007). Ranked lists were im-

ported into the GSEA 4.1.0 and enriched or depleted

pathways were identiﬁed from Gene Ontology: Bi-

ological Processes (GO:BP),(Ashburner et al., 2000)

Reactome,(Wu and Haw, 2017) BioCyc Genome

Database Collection (Karp et al., 2019) and Wikipath-

ways(Martens et al., 2021). Only pathways between

5 and 500 genes were considered.

Second, we used single sample Gene Set Enrich-

ment Analysis (ssGSEA) to reduce the dimension of

each gene expression matrix to a pathway expression

matrix using the ‘GSVA’ package in R (H

anzelmann

et al., 2013). Following above, we only mapped to

pathways between 5 and 500 genes. We then repeated

the ‘limma’ analysis described to identify DEGs using

pathways as biomarker signatures in lieu of genes.

2.3 Replicability of Ranked Lists

The results from the DEGs, GSEA, and ssGSEA

were ranked according in each discovery dataset. To

compare ranked lists, we used Rank Biased Overlap

(RBO). RBO has the beneﬁt of being appropriate for

ranked lists, particularly lists which may contain dif-

ferent numbers of elements and is often used in com-

paring search results (Webber et al., 2010).

2.4 Meta Analysis and Diagnostic Score

Following Sweeney et al. (Sweeney et al., 2015;

Sweeney and Khatri, 2017), we used the meta integra-

tor package (Haynes et al., 2016) on the gene expres-

sion matrices and ssGSEA pathway matrices to con-

duct a random effects meta analysis of possible sepsis

biomarkers in the discovery datasets. Gene/pathway

biomarkers that had a summary FDR p-value < 0.01,

heterogeneity p-value < 0.05, or were not signiﬁ-

cant in all three discovery datasets were ﬁltered out

from further analysis. We ﬁrst considered the ‘top’

biomarker results by taking the 10 biomarkers with

the highest magnitude in summary effect size. We

also constructed a diagnostic score using the greedy

forward based search deﬁned in Sweeney et al.,

wherein the biomarker with the most discriminatory

power is added ﬁrst, and then subsequent biomark-

ers are added based on their improvement to area un-

der the receiving operator characteristics (AUROC)

curves until AUROC no longer improves. This score

is used to construct AUROC curves in both the dis-

covery and validation datasets to compare how gene

or pathway biomarker signatures perform in discrimi-

nating sepsis cases across multiple datasets (Sweeney

et al., 2015; Sweeney and Khatri, 2017; Haynes et al.,

2016).

Replicability of Differentially Expressed Genes Versus Biological Pathways Biomarkers in Diagnosing Sepsis

161

Table 1: The experimental design for the datasets included in the reproducibility analysis.

Split Control deﬁnition Case Deﬁnition n Sample

GSE28750 Validation 24h after ’major surgery’ community-acquired sepsis 21 Blood

GSE32707 Discovery MICU with or without

SIRS, nonseptic

Sepsis, sepsis/ARDS 103 Blood

GSE40012 Discovery SIRS Sepsis from community ac-

quired pneumonia

31 Blood

GSE74224 Discovery Post-surgical infection-

negative systemic inﬂam-

mation

Sepsis patients from ICU 105 Blood

GSE66099 Validation SIRS Sepsis, Septic Shock 229 Blood

Table 2: The RBO results from comparing ranked DEG and pathways from GSEA or ssGSEA.

DEG GSEA ssGSEA FC GSEA FC ssGSEA

GSE32707 vs GSE40012 7.53E-08 4.88E-02 5.45E-07 648 142.48 6.24

GSE32707 vs GSE74224 5.73E-05 1.05E-04 1.99E-04 0.83 2.47

GSE40012 vs GSE74224 7.94E-08 5.43E-02 8.89E-02 683 966.30 1 119 232.86

3 RESULTS AND DISCUSSION

We identiﬁed 5 datasets containing blood samples

from adult patients with sepsis or severe infection.

Three of these datasets were selected for biomarker

discovery, and two were withheld for validation.

Table 1 describes the experimental design of each

dataset. Cases were considered to be sepsis or se-

vere infection, and controls were post-surgical pa-

tients, patients with trauma, and or systemic inﬂam-

matory response syndrome (SIRS).

Complete ranked lists of DEGs were compared

between the three discovery cohorts. The RBO for

each comparison is reported in Table 2. Median RBO

across the three comparisons was 7.94 × 10

−08

. Path-

ways identiﬁed using GSEA were then ranked by

and compared using the RBO metric. The RBO for

the GSEA analysis is also reported in Table 2. Me-

dian RBO across the three GSEA comparisons was

4.88 × 10

−02

. To illustrate the improvement in RBO

similarity, we calculated the fold change improve-

ment in RBO between DEG and GSEA identiﬁed

pathways. As shown in Table 2, the improvement was

large, with two comparisons having a fold change im-

provement of over 6 × 10

While it is unsurprising that ranked lists of

pathways are more reproducible than ranked genes,

canonical methods like GSEA consider pathways at a

dataset level rather than an individual level. Hence,

we should consider whether sample level pathway

analysis is more reproducible than DEGs using the

same analysis techniques that would typically be used

to identify transcriptomic biomarkers.

ssGSEA uses gene expression data to identify a

pathway score for each sample in a data matrix. The

median RBO across our comparisons for ‘differen-

tially expressed’ pathways (DEPs) is 1.99 × 10

−04

with each comparison shown in Table 2. Similar to

the GSEA pathways, all comparisons had consider-

able improvement compared to the DEG ranking.

Despite this improvement in reproducibility, the

RBO was low across both ranked pathways and

ranked genes, reﬂective of biological complexity. It is

also unclear how these improvements in reproducibil-

ity will affect diagnostic performance at the clinical

level. To study this further, we used the meta-analysis

model proposed by Sweeney et al. (Sweeney et al.,

2015; Sweeney and Khatri, 2017) to compare the re-

sults between the gene and pathway signatures.

A critical step of this method is ﬁltering biomark-

ers based on summary effects size, signiﬁcance, and

heterogeneity (Cochran’s Q) (Sweeney et al., 2015;

Sweeney and Khatri, 2017). While ﬁlters based on

the summary effects yielded a similar number of

biomarkers, ﬁltering based on heterogeneity removed

far more genes than pathways. The heterogeneity step

reduced the gene biomarker list to 48 positively and

6 negatively associated genes, conversely, the hetero-

geneity ﬁlter reduced the pathway biomarker list to 55

positive and 50 negative pathways.

The meta analysis results of top 10 associated

genes, by magnitude of effect size, is shown in Figure

1a while the corresponding top pathways are shown in

Figure 1b. The distributions of the summary estimates

BIOINFORMATICS 2022 - 13th International Conference on Bioinformatics Models, Methods and Algorithms

162

Table 3: The genes selected by the greedy forward search in a meta-analysis framework for the diagnosis of sepsis.

Gene FC p-value BH p-value previously associated with sepsis

ADORA2A 1.033 1.63E-06 5.21E-04 Yes (Busse et al., 2016)

ARSD 0.863 9.56E-10 1.64E-06 Yes (Guillen-Guio et al., 2020)

LY6G6D 0.646 3.12E-06 7.78E-04 No

SIGLEC9 0.982 6.14E-12 5.70E-08 Yes (von Gunten et al., 2009)

ZBTB7B 0.648 2.87E-06 7.41E-04 Yes (Bhatty et al., 2012)

TTC17 0.651 2.59E-06 6.96E-04 No

GADD45G 0.643 3.47E-06 8.15E-04 Yes (Aare et al., 2012)

PPP1R9B

0.644 3.32E-06 8.02E-04 No

PARP10 0.634 4.72E-06 9.94E-04 Yes (Wasyluk and Zwolak, 2021)

MPZL2 -0.862 9.69E-10 1.64E-06 Yes (Ji et al., 2014)

DLG5 -0.710 3.36E-07 1.59E-04 Yes (Li et al., 2017)

EPHB4 -0.647 3.14E-06 7.78E-04 Yes (Coulthard et al., 2012)

Table 4: The pathways selected by the greedy forward search in a meta-analysis framework for the diagnosis of sepsis.

Name ID FC p-value BH p-value gene count

Negative regulation of myeloid cell

differentiation

GO:0045638 0.996 3.23E-12 2.82E-08 59

Cell killing GO:0001906 0.723 2.14E-07 6.67E-05 27

Negative regulation of tumor necro-

sis factor superfamily cytokine pro-

duction

GO:1903556 0.955 3.13E-07 8.64E-05 29

Regulation of fatty acid metabolic

process

GO:0019217 -0.698 8.28E-06 7.72E-04 62

APEX1-Independent resolution of

AP sites via the single nucleotide re-

placment pathway

R-HSA-5649702 -0.695 1.26E-05 9.95E-04 7

Polyamine biosyntethic process GO:0006596 -0.624 6.31E-06 6.47E-04 9

Signalling by PTK6 R-HSA-8848021 -0.650 2.72E-06 3.41E-04 61

Regulation of cell cycle G1/S phase

transition

GO:1902806 -0.781 2.45E-08 2.33E-05 110

between the most signiﬁcant genes and pathways is

similar (Kolmogorov-Smirnov p=0.1641), suggesting

that biomarker signatures should be competitive be-

tween genes and pathways.

We identiﬁed a reduced biomarker signature in

both gene expression and pathway matrices using a

greedy forward search algorithm, where biomarkers

were selected based on maximizing the AUC between

cases and controls in the discovery datasets. The ﬁ-

nal meta-score gene signature for sepsis genes can be

found in Table 3 and the meta-score pathway signa-

ture in Table 4.

AUROC for the discovery datasets were promis-

ing, using either the gene meta-score or pathway

meta-score. The AUROCs using the gene meta-

score are 0.815, 0.925, and 0.952 for GSE32707,

GSE40012, and GSE74224 respectively. The AU-

ROCs using the pathway meta-score are 0.833, 0.884,

and 0.925 in GSE32707, GSE40012, and GSE74224

respectively. AUROCs were not signiﬁcantly differ-

ent between the gene signature and the pathway sig-

nature in the discovery datasets.

The AUROC for the validation datasets is shown

in Figure 2. Considerable improvement is seen in the

pathway signature in both datasets. Of note, both

signatures perform poorly in GSE66099, which is

an amalgamation of 6 unique sepsis datasets and in-

cludes patients with septic shock. In particular, the

gene based signature performs only slightly better

than random guessing while the pathway based sig-

nature exhibits poor performance with an AUROC of

0.667. Despite the improvement in a pathway based

signature, more work is needed to increase repro-

ducibility and replicability in validation datasets be-

fore such a signature would be considered for use in a

clinic. However, the improvement of AUROC in the

Replicability of Differentially Expressed Genes Versus Biological Pathways Biomarkers in Diagnosing Sepsis

163

(a) (b)

Figure 1: The top 10 associated genes and pathways from from the ﬁltered meta analysis results.

unique validation datasets does suggest that a path-

way based signature is clinically useful in the devel-

opment of molecular biomarker signatures.

While pathways are more reproducible compared

to genes in this work, much of previous work in

molecular diagnostics has focused on a minimal set

of biomarkers to be measured in order to minimize di-

agnostic costs (Sweeney et al., 2015). However, due

to the falling costs of whole genome RNA sequenc-

ing, we are approaching a time when sequencing the

entire transcriptome will be as cost effective, if not

cheaper, then targeted technologies. Such approaches

open up the possibility of using these pathway based

signatures without additional costs.

One of common critiques of the use of AI in di-

agnosis is a lack of clinical interpretability, with clin-

icians feeling uncomfortable with the ‘black box’ ap-

proach to diagnosis. While pathway signatures cannot

serve to ‘lift the hood’ under complicated AI algo-

rithms, they do allow researchers, clinicians, and pa-

tients to better understand the biological inputs under-

pinning the diagnostic prediction models (Wang et al.,

2020). As the pathways here are generated from the

ranked DEGs, the meta-score pathway signature both

preserves more information that just DEGs while also

increasing interpretability of biomarkers.

Sepsis is characterized by dysregulation in the

host immune system and inﬂammatory response,

which then causes severe oxidative stress.(Macdonald

et al., 2003) Excessive free radicals or inadequate de-

fenses can cause lipid peroxidation, impact cell and

mitochondrial membrane stability and lead to cell

death and tissue damage. (Macdonald et al., 2003;

Fanucchi, 2014) Accordingly, we would expect to

see pathways involved in immune response, heat and

oxidative stress, and apoptosis. Nuclear factor κB

(NFκB) is a ubiquitous transcription factor that is

thought to regulate innate immunity and be involved

in inﬂammation, cancer, and nervous system func-

tion. (Macdonald et al., 2003; Salminen et al., 2008;

Albensi, 2019) By over-regulating downstream pro-

teins, this transcription factor may contribute to the

BIOINFORMATICS 2022 - 13th International Conference on Bioinformatics Models, Methods and Algorithms

164

Figure 2: The gene and pathway meta-score ROC curves in

the validation datasets.

immune dysregulation in sepsis. Several of the path-

ways included in our signature are thought to directly

or indirectly regulate or be regulated by NFκB, in-

cluding: Negative regulation of myeloid cell differen-

tiation (Achyut et al., 2017), Cell killing (Fan et al.,

2008), Negative regulation of tumor necrosis factor

superfamily cytokine production (Hayden and Ghosh,

2014), Regulation of fatty acid metabolism (Kracht

et al., 2020), Polyamine biosynthetic process (Fac-

chini et al., 2005), Regulation of cell cycle G1/S phase

transition (Ledoux and Perkins, 2014). The tight in-

terconnection of these pathways supports the biolog-

ical relevance of our signature, and suggests the rela-

tionship between these processes as a target for future

research in sepsis.

4 CONCLUSION

This work, while promising, is not without limita-

tion. Future work should consider whether the in-

creases seen in reproducibility of biomarkers is true

across many diseases and compare across both mi-

croarray and RNA sequencing. As well, we sought

only to validate these ﬁndings in datasets collected

on the same tissue. Additional work should consider

whether pathway signatures are more reproducible

across different tissue types compared to gene ex-

pression signatures. Moreover, comparing the perfor-

mance of pathway based approaches in different diag-

nostic models should be considered.

We demonstrate that pathway signatures are more

replicable than gene signatures and that pathway sig-

natures can be easily amended to existing signature

identiﬁcation models to improve validation accuracy

in new datasets. Future work emphasizing meth-

ods for pathway-based signatures should occur as

RNA sequencing costs fall and the possibility of cost-

effective pathway signatures becomes reality.

REFERENCES

Aare, S., Radell, P., Eriksson, L. I., Chen, Y.-W., Hoffman,

E. P., and Larsson, L. (2012). Role of sepsis in the

development of limb muscle weakness in a porcine

intensive care unit model. Physiological genomics,

44(18):865–877.

Achyut, B. R., Angara, K., Jain, M., Borin, T. F., Rashid,

M. H., Iskander, A. S. M., Ara, R., Kolhe, R., Howard,

S., Venugopal, N., Rodriguez, P. C., Bradford, J. W.,

and Arbab, A. S. (2017). Canonical nfκb signaling in

myeloid cells is required for the glioblastoma growth.

Scientiﬁc Reports, 7(1):13754.

Albensi, B. C. (2019). What is nuclear factor kappa b (nf-

κb) doing in and to the mitochondrion? Frontiers in

Cell and Developmental Biology, 7:154.

Alpern, D., Gardeux, V., Russeil, J., Mangeat, B., Meireles-

Filho, A. C. A., Breysse, R., Hacker, D., and De-

plancke, B. (2019). BRB-seq: ultra-affordable high-

throughput transcriptomics enabled by bulk RNA bar-

coding and sequencing. Genome Biology, 20(1):71.

Ashburner, M., Ball, C. A., Blake, J. A., Botstein, D.,

Butler, H., Cherry, J. M., Davis, A. P., Dolinski, K.,

Dwight, S. S., Eppig, J. T., Harris, M. A., Hill, D. P.,

Issel-Tarver, L., Kasarskis, A., Lewis, S., Matese,

J. C., Richardson, J. E., Ringwald, M., Rubin, G. M.,

and Sherlock, G. (2000). Gene Ontology: tool for the

uniﬁcation of biology. Nature Genetics, 25(1):25–29.

Barrett, T., Wilhite, S. E., Ledoux, P., Evangelista, C., Kim,

I. F., Tomashevsky, M., Marshall, K. A., Phillippy,

K. H., Sherman, P. M., Holko, M., Yefanov, A., Lee,

H., Zhang, N., Robertson, C. L., Serova, N., Davis,

S., and Soboleva, A. (2012). NCBI GEO: archive for

functional genomics data sets—update. Nucleic Acids

Research, 41(D1):D991–D995.

Bhatty, M., Fan, R., Muir, W. M., Pruett, S. B., and Nan-

duri, B. (2012). Transcriptomic analysis of peritoneal

cells in a mouse model of sepsis: conﬁrmatory and

novel results in early and late sepsis. BMC genomics,

13(1):1–13.

Busse, H., Bitzinger, D., H

ocherl, K., Seyfried, T., Gruber,

M., Graf, B. M., and Zausig, Y. A. (2016). Adeno-

sine a2a and a2b receptor substantially attenuate is-

chemia/reperfusion injury in septic rat hearts. Cardio-

vascular drugs and therapy, 30(6):551–558.

Chen, J. J., Tsai, C.-A., Tzeng, S., and Chen, C.-H. (2007).

Gene selection with multiple ordering criteria. BMC

Bioinformatics, 8(1):74.

Cohen, J., Vincent, J.-L., Adhikari, N. K. J., Machado, F. R.,

Angus, D. C., Calandra, T., Jaton, K., Giulieri, S., De-

laloye, J., Opal, S., Tracey, K., van der Poll, T., and

Replicability of Differentially Expressed Genes Versus Biological Pathways Biomarkers in Diagnosing Sepsis

165

Pelfrene, E. (2015). Sepsis: a roadmap for future re-

search. The Lancet Infectious Diseases, 15(5):581–

614.

Coulthard, M. G., Morgan, M., Woodruff, T. M., Aru-

mugam, T. V., Taylor, S. M., Carpenter, T. C., Lack-

mann, M., and Boyd, A. W. (2012). Eph/ephrin signal-

ing in injury and inﬂammation. The American journal

of pathology, 181(5):1493–1503.

Crow, M., Lim, N., Ballouz, S., Pavlidis, P., and Gillis, J.

(2019). Predictability of human differential gene ex-

pression. Proceedings of the National Academy of Sci-

ences of the United States of America, 116(13):6491–

6500.

Dolinay, T., Kim, Y. S., Howrylak, J., Hunninghake,

G. M., An, C. H., Fredenburgh, L., Massaro, A. F.,

Rogers, A., Gazourian, L., Nakahira, K., Haspel,

J. A., Landazury, R., Eppanapally, S., Christie, J. D.,

Meyer, N. J., Ware, L. B., Christiani, D. C., Ry-

ter, S. W., Baron, R. M., and Choi, A. M. K.

(2012). Inﬂammasome-regulated Cytokines Are Crit-

ical Mediators of Acute Lung Injury. American

Journal of Respiratory and Critical Care Medicine,

185(11):1225–1234.

Facchini, A., Borz

I, R. M., Marcu, K. B., Stefanelli, C.,

Olivotto, E., Goldring, M. B., Facchini, A., and

Flamigni, F. (2005). Polyamine depletion inhibits

nf-κb binding to dna and interleukin-8 production

in human chondrocytes stimulated by tumor necrosis

factor-α. Journal of Cellular Physiology, 204(3):956–

963.

Fan, Y., Dutta, J., Gupta, N., Fan, G., and G

elinas, C.

(2008). Regulation of programmed cell death by nf-

κb and its role in tumorigenesis and therapy. In Pro-

grammed Cell Death in Cancer Progression and Ther-

apy, volume 615, pages 223–250. Springer Nether-

lands, Dordrecht. Series Title: Advances in Experi-

mental Medicine and Biology.

Fanucchi, M. V. (2014). Development of antioxidant and

xenobiotic metabolizing enzyme systems.

Ferrer, R., Martin-Loeches, I., Phillips, G., Osborn, T. M.,

Townsend, S., Dellinger, R. P., Artigas, A., Schorr,

C., and Levy, M. M. (2014). Empiric Antibiotic

Treatment Reduces Mortality in Severe Sepsis and

Septic Shock From the First Hour: Results From

a Guideline-Based Performance Improvement Pro-

gram*. Critical Care Medicine, 42(8):1749–1755.

Guillen-Guio, B., Lorenzo-Salazar, J. M., Ma, S.-F., Hou,

P.-C., Hernandez-Beeftink, T., Corrales, A., Garc

ıa-

Laorden, M. I., Jou, J., Espinosa, E., Muriel, A.,

et al. (2020). Sepsis-associated acute respiratory dis-

tress syndrome in individuals of european ancestry: a

genome-wide association study. The Lancet Respira-

tory Medicine, 8(3):258–266.

anzelmann, S., Castelo, R., and Guinney, J. (2013).

GSVA: gene set variation analysis for microarray and

RNA-Seq data. BMC Bioinformatics, 14:7.

Hayden, M. S. and Ghosh, S. (2014). Regulation of nf-

κb by tnf family cytokines. Seminars in Immunology,

26(3):253–266.

Haynes, W. A., Vallania, F., Liu, C., Bongen, E., Tomczak,

A., Andres-Terre, M., Lofgren, S., Tam, A., Deis-

seroth, C. A., Li, M. D., Sweeney, T. E., and Khatri,

P. (2016). Empowering multi-cohort gene expression

analysis to increase reproducibility. Paciﬁc Sympo-

sium on Biocomputing.

Ji, S., Pan, Y., Lu, Q., Sun, Z., and Liu, Y. (2014). Screen-

ing of differentially expressed genes between multiple

trauma patients with and without sepsis. Genet Mol

Res, 13(1):1855–64.

Karp, P. D., Billington, R., Caspi, R., Fulcher, C. A., La-

tendresse, M., Kothari, A., Keseler, I. M., Krummen-

acker, M., Midford, P. E., Ong, Q., Ong, W. K., Paley,

S. M., and Subhraveti, P. (2019). The BioCyc collec-

tion of microbial genomes and metabolic pathways.

Brieﬁngs in Bioinformatics, 20(4):1085–1093.

Khatri, P., Sirota, M., and Butte, A. J. (2012). Ten Years

of Pathway Analysis: Current Approaches and Out-

standing Challenges. PLoS Computational Biology,

8(2):e1002375.

Kim, K. S., Jekarl, D. W., Yoo, J., Lee, S., Kim, M., and

Kim, Y. (2021). Immune gene expression networks

in sepsis: A network biology approach. PLOS ONE,

16(3):e0247669.

Kracht, M., M

uller-Ladner, U., and Schmitz, M. L. (2020).

Mutual regulation of metabolic processes and proin-

ﬂammatory nf-κb signaling. Journal of Allergy and

Clinical Immunology, 146(4):694–705.

Ledoux, A. and Perkins, N. (2014). Nf-κb and the cell cy-

cle. Biochemical Society Transactions, 42(1):76–81.

Li, Z., Zhang, Y., Liu, Y., Liu, Y., and Li, Y. (2017). Iden-

tiﬁcation of key genes in gram-positive and gram-

negative sepsis using stochastic perturbation. Molec-

ular medicine reports, 16(3):3133–3146.

Lim, K., Li, Z., Choi, K. P., and Wong, L. (2015). A quan-

tum leap in the reproducibility, precision, and sensi-

tivity of gene expression proﬁle analysis even when

sample size is extremely small. Journal of Bioinfor-

matics and Computational Biology, 13(04):1550018.

Macdonald, J., Galley, H., and Webster, N. (2003). Ox-

idative stress and gene expression in sepsis. British

Journal of Anaesthesia, 90(2):221–232.

Martens, M., Ammar, A., Riutta, A., Waagmeester, A.,

Slenter, D., Hanspers, K., A. Miller, R., Digles, D.,

Lopes, E., Ehrhart, F., Dupuis, L. J., Winckers, L. A.,

Coort, S., Willighagen, E. L., Evelo, C. T., Pico, A. R.,

and Kutmon, M. (2021). WikiPathways: connecting

communities. Nucleic Acids Research, 49(D1):D613–

D621.

Mayday, M. Y., Khan, L. M., Chow, E. D., Zinter, M. S., and

DeRisi, J. L. (2019). Miniaturization and optimiza-

tion of 384-well compatible RNA sequencing library

preparation. PLOS ONE, 14(1):e0206194.

McHugh, L., Seldon, T. A., Brandon, R. A., Kirk, J. T.,

Rapisarda, A., Sutherland, A. J., Presneill, J. J., Ven-

ter, D. J., Lipman, J., Thomas, M. R., Klein Klouwen-

berg, P. M. C., van Vught, L., Scicluna, B., Bon-

ten, M., Cremer, O. L., Schultz, M. J., van der Poll,

T., Yager, T. D., and Brandon, R. B. (2015). A

Molecular Host Response Assay to Discriminate Be-

tween Sepsis and Infection-Negative Systemic Inﬂam-

mation in Critically Ill Patients: Discovery and Val-

BIOINFORMATICS 2022 - 13th International Conference on Bioinformatics Models, Methods and Algorithms

166

idation in Independent Cohorts. PLOS Medicine,

12(12):e1001916.

Mootha, V. K., Lindgren, C. M., Eriksson, K.-F., Subrama-

nian, A., Sihag, S., Lehar, J., Puigserver, P., Carls-

son, E., Ridderstr

ale, M., Laurila, E., Houstis, N.,

Daly, M. J., Patterson, N., Mesirov, J. P., Golub,

T. R., Tamayo, P., Spiegelman, B., Lander, E. S.,

Hirschhorn, J. N., Altshuler, D., and Groop, L. C.

(2003). PGC-1α-responsive genes involved in oxida-

tive phosphorylation are coordinately downregulated

in human diabetes. Nature Genetics, 34(3):267–273.

Parnell, G. P., McLean, A. S., Booth, D. R., Armstrong,

N. J., Nalos, M., Huang, S. J., Manak, J., Tang, W.,

Tam, O.-Y., Chan, S., and Tang, B. M. (2012). A

distinct inﬂuenza infection signature in the blood tran-

scriptome of patients with severe community-acquired

pneumonia. Critical Care, 16(4):R157.

Pradines, J. R., Farutin, V., Cilfone, N. A., Ghavami, A.,

Kurtagic, E., Guess, J., Manning, A. M., and Capila,

I. (2020). Enhancing reproducibility of gene expres-

sion analysis with known protein functional relation-

ships: The concept of well-associated protein. PLOS

Computational Biology, 16(2):e1007684.

R Core Team (2017). R: A Language and Environment for

Statistical Computing. R Foundation for Statistical

Computing, Vienna, Austria.

Ritchie, M. E., Phipson, B., Wu, D., Hu, Y., Law, C. W., Shi,

W., and Smyth, G. K. (2015). limma powers differen-

tial expression analyses for RNA-sequencing and mi-

croarray studies. Nucleic Acids Research, 43(7):e47–

e47.

Salminen, A., Huuskonen, J., Ojala, J., Kauppinen, A.,

Kaarniranta, K., and Suuronen, T. (2008). Activation

of innate immunity system during aging: NF-kB sig-

naling is the molecular culprit of inﬂamm-aging. Age-

ing Research Reviews, 7(2):83–105.

Scicluna, B. P., Klein Klouwenberg, P. M. C., van Vught,

L. A., Wiewel, M. A., Ong, D. S. Y., Zwinderman,

A. H., Franitza, M., Toliat, M. R., N

urnberg, P.,

Hoogendijk, A. J., Horn, J., Cremer, O. L., Schultz,

M. J., Bonten, M. J., and van der Poll, T. (2015).

A Molecular Biomarker to Diagnose Community-

acquired Pneumonia on Intensive Care Unit Admis-

sion. American Journal of Respiratory and Critical

Care Medicine, 192(7):826–835.

Sholder, G., Lanz, T. A., Moccia, R., Quan, J., Aparicio-

Prat, E., Stanton, R., and Xi, H. S. (2020). 3’Pool-seq:

an optimized cost-efﬁcient and scalable method of

whole-transcriptome gene expression proﬁling. BMC

Genomics, 21(1):64.

Subramanian, A., Tamayo, P., Mootha, V. K., Mukher-

jee, S., Ebert, B. L., Gillette, M. A., Paulovich, A.,

Pomeroy, S. L., Golub, T. R., Lander, E. S., and

Mesirov, J. P. (2005). Gene set enrichment analysis: A

knowledge-based approach for interpreting genome-

wide expression proﬁles. Proceedings of the National

Academy of Sciences, 102(43):15545–15550.

Sutherland, A., Thomas, M., Brandon, R. A., Brandon,

R. B., Lipman, J., Tang, B., McLean, A., Pascoe,

R., Price, G., Nguyen, T., Stone, G., and Venter, D.

(2011). Development and validation of a novel molec-

ular biomarker diagnostic test for the early detection

of sepsis. Critical Care, 15(3):R149.

Sweeney, T. E. and Khatri, P. (2017). Benchmarking Sep-

sis Gene Expression Diagnostics Using Public Data*:.

Critical Care Medicine, 45(1):1–10.

Sweeney, T. E., Shidham, A., Wong, H. R., and Khatri, P.

(2015). A comprehensive time-course–based multico-

hort analysis of sepsis and sterile inﬂammation reveals

a robust diagnostic gene set. Science Translational

Medicine, 7(287):287ra71–287ra71.

Tan, P. K. (2003). Evaluation of gene expression measure-

ments from commercial microarray platforms. Nu-

cleic Acids Research, 31(19):5676–5684.

Tweedie, S., Braschi, B., Gray, K., Jones, T. E. M., Seal, R.,

Yates, B., and Bruford, E. A. (2021). Genenames.org:

the HGNC and VGNC resources in 2021. Nucleic

Acids Research, 49(D1):D939–D946.

van Zanten, A. R. H. (2014). The Golden Hour of Antibi-

otic Administration in Severe Sepsis: Avoid a False

Start Striving for Gold*. Critical Care Medicine,

42(8):1931–1932.

von Gunten, S., Jakob, S. M., Geering, B., Takala, J., and

Simon, H.-U. (2009). Different patterns of siglec-9-

mediated neutrophil death responses in septic shock.

Shock, 32(4):386–392.

Wang, F., Kaushal, R., and Khullar, D. (2020). Should

health care demand interpretable artiﬁcial intelligence

or accept “black box” medicine?

Wasyluk, W. and Zwolak, A. (2021). Parp inhibitors: An

innovative approach to the treatment of inﬂammation

and metabolic disorders in sepsis. Journal of Inﬂam-

mation Research, 14:1827.

Webber, W., Moffat, A., and Zobel, J. (2010). A similarity

measure for indeﬁnite rankings. ACM Transactions on

Information Systems, 28(4):1–38.

Wu, G. and Haw, R. (2017). Functional Interaction Net-

work Construction and Analysis for Disease Discov-

ery. In Wu, C. H., Arighi, C. N., and Ross, K. E.,

editors, Protein Bioinformatics, volume 1558, pages

235–253. Springer New York, New York, NY. Series

Title: Methods in Molecular Biology.

Zarringhalam, K., Enayetallah, A., Reddy, P., and Ziemek,

D. (2014). Robust clinical outcome prediction based

on Bayesian analysis of transcriptional proﬁles and

prior causal networks. Bioinformatics, 30(12):i69–

i77.

Zhang, M., Yao, C., Guo, Z., Zou, J., Zhang, L., Xiao, H.,

Wang, D., Yang, D., Gong, X., Zhu, J., Li, Y., and Li,

X. (2008). Apparently low reproducibility of true dif-

ferential expression discoveries in microarray studies.

Bioinformatics, 24(18):2057–2063.

Zhang, M., Zhang, L., Zou, J., Yao, C., Xiao, H., Liu,

Q., Wang, J., Wang, D., Wang, C., and Guo, Z.

(2009). Evaluating reproducibility of differential ex-

pression discoveries in microarray studies by consid-

ering correlated molecular changes. Bioinformatics,

25(13):1662–1668.

Replicability of Differentially Expressed Genes Versus Biological Pathways Biomarkers in Diagnosing Sepsis

167