Abstract
A DEEP LEARNING METHOD UNCOVERS NOVEL PATHWAY ASSOCIATIONS IN SYSTEMIC LUPUS ERYTHEMATOSUS
Full text
Background: While there have been recent improvements in therapy for systemic lupus erythematosus (SLE), unmet need remains for novel, efficacious treatments. Artificial intelligence can facilitate drug discovery by identifying potential target molecules and pathways.
Objectives: To devise a Deep Learning method that ranks genes and pathways differentially expressed in SLE by an integrated metric of expression and interactivity.
Methods: HT2A array expression data (17,799 transcripts) were obtained from whole blood samples taken at baseline (N=1,760) from patients with SLE enrolled in two phase 3 trials (NCT01196091 and NCT01205438). We applied a Deep learning algorithm
to create a disease-specific gene network for SLE from the Decagon and MetaCore molecular interaction databases
. A meta-analysis was conducted for each candidate gene by calculating its distance from other genes in the disease-specific network (interactivity analyses). The distance was then multiplied by each gene’s utility score, a measure of the gene’s relevance to SLE disease activity derived from public datasets
. The final output was the fractional rank (0–1) of utilities
for each gene against prespecified outcomes of SLEDAI-2K and IFN gene signatures, analyzed as a continuous variable combining 34 IFN-related genes. Selected JAK-STAT, cytokine, and IFN genes are presented. Pathways enriched in the top 250 genes from the input data and Deep Learning output were analyzed by hypergeometric tests.
Results: The highest-ranked gene for SLEDAI-2K and IFN signature status was IFNγ, followed, as expected, by the IFNα gene family (
Table 1
). Application of the Deep Learning method markedly affected pathway analysis outcomes, revealing strong associations in pathways including those related to CD4+ T cell differentiation as well as macrophage and dendritic cells, which were not apparent in conventional analysis (
Figure 1
).
Table 1.
Gene
SLEDAI-2K fractional rank
IFN signature fractional rank
IFNγ
0.9999
0.9999
IFNα†
0.999*
0.999*
STAT1
0.9991
0.9991
IL10
0.9990
0.9987
IFNB1
0.9990
0.9989
STAT6
0.9981
0.9966
IL6
0.9974
0.9973
STAT4
0.9959
0.9944
IL12A
0.9952
0.9932
IL12B
0.9950
0.9938
STAT5A
0.9929
0.9940
STAT3
0.9915
0.9930
IFNγR2
0.9829
0.9770
IFNγR1
0.9812
0.9840
IL23A
0.9806
0.9895
IL10RA
0.9805
0.9816
IL6ST
0.9791
0.9695
STAT2
0.9772
0.9914
†IFNα genes were similarly ranked.
*Range: 0.9988–0.9999.
“0.9991” denotes that the SLEDAI-2K rank for STAT1 is higher than 99.91% of the whole genome.
Conclusion: Deep learning algorithms incorporating interactivity analyses revealed disease associations in SLE that were not apparent when using conventional bioinformatic approaches.
REFERENCES:
[1]Kong Y and Yu T. Bioinformatics. 2018;34(21):3727–37.
[2]
http://snap.stanford.edu/decagon/
.
[3]
https://portal.genego.com/
.
[4]Valdeolivas A et al. Bioinformatics. 2019;35(3):497–505.
[5]
https://cran.r-project.org/web/packages/MetaIntegrator/index.html
.
Figure 1.
Disclosure of Interests: Eric F. Morand Speakers bureau: AstraZeneca, Paid instructor for: Eli Lilly, Consultant of: Amgen, AstraZeneca, Biogen, BristolMyersSquibb, Eli Lilly, EMD Serono, Genentech, GlaxoSmithKline, Janssen, Grant/research support from: AstraZeneca, BristolMyersSquibb, Eli Lilly, EMD Serono, GlaxoSmithKline, Janssen, Thomas Dörner Speakers bureau: Eli Lilly, Samsung, Roche, Consultant of: AbbVie, Celgene, Eli Lilly, Janssen, Novartis, Roche, Samsung, UCB, Grant/research support from: Janssen, Novartis, Roche, UCB, Sanofi, Bochao Jia Shareholder of: Eli Lilly, Employee of: Eli Lilly, Yushi Liu Shareholder of: Eli Lilly, Employee of: Eli Lilly, Stephanie de Bono Shareholder of: Eli Lilly, Employee of: Eli Lilly, Damiano Fantini Shareholder of: Eli Lilly, Employee of: Eli Lilly, Maria Silk Shareholder of: Eli Lilly, Employee of: Eli Lilly, Natalia Bello Vega Shareholder of: Eli Lilly, Employee of: Eli Lilly, Peter Fischer Shareholder of: Eli Lilly, Employee of: Eli Lilly, Michelle A Petri Consultant of: Eli Lilly, Grant/research support from: Eli Lilly
Citation: Ann Rheum Dis, volume 80, supplement 1, year 2021, page 406Session: Genomics, genetic basis of disease and functional genomics
(POSTERS only)
4 organizations