(197f) From Experimental Data to Mechanistic Hypotheses: Analysis of Proteomic Data Using a Very Large-Scale Causal Model

High-throughput proteomic analyses of tissue and bio-fluid samples can yield datasets comprising measured differences in hundreds - or even thousands - of proteins. In principle, this rich source of data can provide a systems-level view of the biological processes in an experiment, leading to testable hypotheses describing the mechanisms that led to the observed changes. But typically, the integration of hundreds of observations to infer the active biological networks is an unmanageable task, limiting the analysis to categorization of the changed proteins by annotations and by patterns of modulation. To identify disease mechanisms, compound mechanisms and biomarkers from proteomic and systems biology experiments requires the development of a model of biology. Using a mental model, a scientist can reason about hundreds of distinct molecules present within a cell, but reasoning over tens of thousands of molecules and their interrelationships is impossible. We describe the development and application of a very large-scale causal, computable model of biology which has been used to identify molecular cause and effect hypotheses consistent with data from proteomic experiments. Automated causal analysis can be used to define upstream networks of molecular events which could result in experimentally observed protein changes. It can be used to identify possible causal pathways linking initial experimental perturbations to observed protein or phenotypic changes. Large-scale causal analysis is a powerful new systems-based approach for the interpretation of molecular state measurements in drug discovery.