Download Gene Set Analysis with Phenotypic Screening Data Results and Validation Purpose

Gene Set Analysis with Phenotypic Screening Data
Charles Hoyt, Elisabet Gregori-Puigjane, Miguel Camargo
Center for Proteomic Chemistry, Novartis Institutes for Biomedical Research, Cambridge, Massachusetts
Results and Validation
Compound
Purpose
Down-regulated Tail
• Identify pathways relevant to a phenotypic
screen
• Elucidate active compounds’ mechanisms of
action in a phenotypic screen
• Find new targets by expanding around
validated targets in relevant gene sets
in silico hypothesis
to link the
phenotype to a
compounds
Up-regulated Tail
Target-centric
lead discovery
Disease
Pathway
Phenotype
Genetics
in silico hypothesis to link
the phenotype to a MoA
Target
Background
• Most gene set analysis methods were
developed for gene-centric data
• Of these methods, many focus on genes with
extreme readouts, and can overlook
significant aggregate effects due to many
genes with more subtle readouts
• Pathway Influence Scoring was developed to
address that issue for gene-centric data
• The method was then adapted for use with
compound data
• The plot shows scores versus the p-values to help distinguish significant
gene sets that are up-regulated versus down-regulated
• Sensitivity analyses of the net and absolute methods have been
conducted to measure the robustness of the techniques and detect false
positive gene sets
• The analysis was run on a viral infection cell proliferation assay then the
significant sets were clustered (below). The themes are consistent with
validated targets and pathways in viral infection.
Challenges
• Using small molecules as probes faces its
own challenges.
• Their target annotations are less complete
than the siRNA/related probes and may be
acting through an unknown mechanism.
• The molecules might also lack the ability to
enter the cell or get to their intended targets.
• This is addressed by using larger screens
with highly annotated compounds.
Methods
• Compounds are annotated to genes for which they have an IC50 < 10μM
• Genes’ activities are calculated by aggregating their annotated compounds’
phenotypic activities with an arithmetic mean
Readout 1
Compound 1
Readout 2
Compound 2
Readout 3
Compound 3
Gene A
• Each gene set is scored using a net and absolute scheme
• P-values are calculated for each set and method using bootstrapping – the original set’s
score was compared to the scores of 10000 sets built from genes randomly sampled
from the population
A. Transglutiminase Pathways
B. Immune Response, PI3K
Pathways
C. Surface Receptors and Ion
Channels
D. Cellular Metabolism
E. Viral Life Cycle
F. Tetratricopeptide Domain
Conclusions
• The method has been validated and is sufficiently robust
• New cell-based assays can be analyzed
Looking Forwards
• Comparison with adaptations of Broad’s GSEA method and other
statistical methods
• Additional post-processing techniques, including the contextualization
of results with literature and text-mining approaches
Acknowledgements
Special Thanks to: Elisabet Gregori-Puigjane, Miguel Camargo, Stephan
Reiling, Shreyas Mahimkar, and everyone in iSLD and CPC.