Genstruct - Genstruct Methodology

Genstruct Methodology

Hypothesis-Driven Research

Molecular profiling has been limited in its use as a research discovery tool because it has thus far only delivered measurement rather than hypotheses. This has been due to a lack of process to translate high throughput profiling data into a coherent framework of scientific knowledge.

Through the use of computational reasoning and inference, Genstruct’s methodology transforms profiling experiments into hypotheses generating experiments by building causal models of biology that can be explored against profiling data.

The Genstruct methodology begins with the Knowledge Assembly™ process, which produces a Causal System Model specific to each theraputic question at hand. The model is defined through Genstruct’s Causal Reasoning and Inferencing process (illustrated in the figure below), which involves interative model testing, causal analysis and hypothesis generation.

The final product is a coherent system view of the perturbed biology as implicated by experimental data, and the resulting Causal System Model is then interrogated by analytics to identify mechanisms of action for diseases and drugs, as well as to deliver mechanistic biomarkers.

Knowledge Assembly™ Process
The process of knowledge assembly begins by developing a high-level concept model of the biological system. This concept model defines the cellular processes and their inter-relationships that comprise the biology of the system, and defines the genes, proteins and metabolites present in the system. The process continues by drawing from the full Knowledge Repository, which includes all of the causal facts that have ever been encoded into the system about each of the genes, proteins, metabolites, cells, tissues and concepts needed to build the model of the system. Finally, each of the biological state changes that will be used to develop the Causal System Model are rigorously researched to identify their causal relationships to the components of the model, in order to facilitate causal reasoning about the control of their states. As the modeling nears completion, a suite of tests are developed and run to ensure that the model accurately reproduces the known biology of the system. At this point, the model can be used for causal reasoning.

Biological State Change Analysis
Experimental data are continuous, mathematical measurements that must be converted into causal changes of biological states to drive causal reasoning. To accomplish this, Genstruct analyzes the experimental data to identify a set of statistically significant biological signals that define the change in state induced by the disease or drug treatment. Data quality is incredibly important, and each signal must pass stringent quality controls to ensure that they are sufficient to drive causal reasoning. The biological states that we focus on include changes in transcriptional regulation, protein phosphorylation, protein processing, protein degradation, receptor activation, metabolic conversion and more. Each of these states are then connected to their regulatory machinery as represented in the knowledge assembly.

Iterative Refinement
Each of the biological state changes identified in our analysis has a multitude of possible control mechanisms. We employ reverse causal reasoning to identify the potential upstream regulators of these state changes. Each of the regulators identified in reverse causal reasoning will also control other biological states, and each of these are evaluated through a forward causal reasoning. This process results in millions of potential chains (hypotheses) that can explain aspects of the data. These hypotheses are scored and evaluated using our proprietary methods, to define the most biologically significant networks as defined by the data, the model and the reasoning method. These hypotheses are explored further through additional literature research, remodeled, and each hypothesis is either refuted, or supported and further refined.

Causal System Modeling

The ability to capture the activated networks and pathways of a biological system perturbed by disease or drug into a single model has never been possible before. Genstruct has developed the method for identifying the appropriate variations of each perturbed subsystem into a single, coherent model that can be evaluated and simulated to evaluate potential targets, to identify potential biomarkers and to simply ask “what if” questions. The resulting causal system models can typically correctly explain more than 70% of the observed biological state changes in an experiment, and contain all the potential intervention points for and markers of the biology that has been defined. These models are developed for each program that we undertake with our partners and independently, and represent a causal framework that can integrate all available discovery knowledge into a single, compact knowledge representation.