Design principles for clinical network-based proteomics

• Design principles for network-based proteomics are lacking.
• Advances in statistics, network biology and proteomics drive each other reciprocally.
• Methodology development must address statistical, methodological and biological rigor.
• Less-well-appreciated considerations include normalization issues and test inflation.

Integrating biological networks with proteomics is a tantalizing option for system-level analysis; for example it can help remove false-positives from proteomics data and improve coverage by detecting false-negatives, as well as resolving inconsistent inter-sample protein expression due to biological heterogeneity. Yet, designing a robust network-based analysis strategy on proteomics data is nontrivial. The issues include dealing with test set bias caused by, for example, inappropriate normalization procedure, devising appropriate benchmarking criteria and formulating statistically robust feature-selection techniques. Given the increasing importance of proteomics in contemporary clinical studies, more powerful network-based approaches are needed. We provide some design principles and considerations that can help achieve this, while taking into account the idiosyncrasies of proteomics data.