This document summarizes an ABRF workshop on pathway analysis in genomics, transcriptomics, proteomics, and metabolomics. It discusses that pathway analysis (PA) techniques are needed to understand large amounts of "omics" data in the context of biological systems. PA methods take lists of genes and produce reduced lists and models. The document outlines different types of PA tools, issues with training, assumptions that must be verified, and knowledge bases that are used. It also notes limitations of current approaches and the need for new developments like better software, data standards, and validation strategies.

1. ABRF Workshop
Pathway Analysis in
Genomics, Transcriptomics,
Proteomics and
Metabolomics
Orlando, FL 2012/03/17

2. Why PA now?
Driven by the need to understand mechanisms
in Biology
Fed by a torrent of “omics” data
Integrated in Systems Biology approaches
Towards P4 Medicine, environmental, pharma

3. What is PA
A set of techniques that provide insight into the
structure and functioning of biological systems,
allowing for inference on their behavior.
The methods start from large lists of genes and
their annotation, use quantitative levels of
expression and produce reduced lists and/or
models.

4. Training issues
Three day long training courses in Oeiras, PT
For about 20 people
Connected to other themes:
Gene Regulatory Networks (2009)
Chemoinformatics (2010)
Network Biology (2011)
Results: need more time, more statistics bkgnd

5. Well conducted PA will assume
Verification of the assumptions
Statistics proficiency (trained users)
High quality of the annotations
We need more and better knowledge bases
Abundance of omics data
More (open) data from tech services, targeted research

6. Basics
● From a list of genes from HT experiments in
two different conditions, we want to be able to
reliably extract reduced lists of genes that
justify the functional change.
● We also want to interpret the above findings in
the context of biological mechanisms being
present/absent.

7. Knowledge bases
Provide annotated gene sets
KEGG
REACTOME
METACYC
GO

8. Types of PA tools
ORA (Over-Representation Analysis) ex: FatiGo
FCS (Functional Class Scoring) ex: segPathway
PT (Pathway Topology) ex : GenExplain
Reference
Ten Years of Pathway Analysis: Current Approaches and Outstanding Challenges
Purvesh Khatri, Marina Sirota, Atul J. Butte
PLOS Computational Biology, Feb 2012, vol. 8, Issue 2, e1002375

9. ORA Methods
Works on counts of genes
Ignores signal “strength”
Filters-out by thresholding
Assumes independence between pathways

10. FCS approaches
Statistical classification and grouping
Correlation with phenotypic characters
Significance test between Pw statistics

11. PT-based techniques
Like FCS, but make good use of co-expression
and interaction annotations
Single-out topological models that explain the
data

12. All the above
May produce defective results
In presence weak annotation quality
Failure to meet statistical assumptions
Artifacts

13. However...
We need to be ready for upcoming data
deluges and their consequences
Medical information coming from new corners
- adverse effects database (Stanford)
- patients data collections

14. New developments
We need
● a significant investment in software
development
● data interoperability and standards for cross-
platform exchange
● better validation strategies

15. We need trained users