2. Metabolomics study
• Metabolomics study
• Identify and quantify metabolites/correlate their changes with pathological
states or external influencing factors
* Image from Davis VW et al. J Surg Oncol. (2011)
3. Metabolomics study
• How to interpret metabolomics data?
• Metabolic pathway analysis
• network-based visualization
• The multi-level integration is
fundamental!
• However, challenges in complexity and
heterogeneity
MetaboAnalyst (https://www.metaboanalyst.ca/)
4. Metabolomics study
• Targeted metabolomics
• Detection and precise quantification (in nM, or mg/mL) of a small set of
known compounds
• Need to clearly define “standards”
• Untargeted metabolomics (metabolite fingerprinting)
• “complete” metabolome comparison (many metabolites as possible)
• Technical limitations, bias toward the most abundant molecules
• How to handle unknown metabolites?
8. PIUMet
3. Infer a subnetwork connecting disease
features using Prize-collecting Steiner
forest (PCSF) optimization
• Maximize the sum of prizes (significance of
dysregulation) from connected disease features
• Minimize the edge costs (anti-correlated with
edge confidence)
• disease features ↑, low-confidence edges ↓
Node penalties Edge costs
Minimize
Input The possible PCSF solution
9. PIUMet
4. Eliminate bias for highly connected nodes by assigning a penalty
5. Merge inferred family of networks from many runs by adding a
random noise to capture metabolic network complexity
6. Calculate disease-specific scores for each node and a network by
generating networks from random data mimicking the experimental
data
10. Experiments in HD
115 metabolite
features
P
31 proteins
m/z
M M M
37 metabolite peaks
(disease feature)
296 metabolites
PIUMet can detect many disease features
compared to targeted ones
13. Integrating with other omics
Integrative analysis can detect many
novel disease features
Node scores increase with
a joint analysis of multi-omics
14. Conclusion
• Challenges in global metabolite identification in untargeted
metabolomics
• PIUMet, a network-based PCSF algorithm for integrative analysis of
untargeted metabolomics
• Identify novel disease-associated metabolites and proteins that
cannot be found in individual data using PPMI network
• No prior assumptions, it can be generally used in other disease data