Method to Determine Causal Relationships in Complex Networks

Background

As the volume of large-scale data generated in both research and commercial settings rapidly expands, accessing key insight on causal connections in a reliable and quick manner is increasingly crucial. Computational tools that accomplish this task will be useful in applications ranging from accurately modeling complex diseases such as cancer and Alzheimer’s disease to predicting how stock prices influence each other. However, currently available methods of analysis are restrictive, as they require individual components to be of similar type and fail to resolve causal relationships accurately while establishing association. In addition, they require integration of additional data in order to untangle causality.

Technology Overview

This mathematical method identifies causal relationships between multiple variables in complex networks of high-throughput datasets.

Using a unique algorithmic approach, this technology accurately infers causal relationships in multivariate datasets without a need to integrate additional data. This is particularly useful where additional datasets are unavailable or the dataset is underpowered. This algorithm has been validated in multiple types of datasets for biological research and financial applications. For example, it successfully identified causal connections among genes expressed in a metabolic network. It also determined the causal relationship between the stock price of an oil company and a retail store, specifically detailing how the price of the oil company’s stock impacted the stock price of the retail store and not vice-versa. The technology has a high commercial potential due to its versatile application in large-scale, multivariate datasets without the need for prior training.

Stage of Development

• Research use

Further Details

• Chang, R; Karr, J; Schadt, E. Causal Inference in Biological Networks with Integrated Belief Propagation. http://psb.stanford.edu/psb-online/proceedings/psb15/chang.pdf
• Schadt, E. et al An Integrative Genetic Approach to Infer Causal Associations Between Gene Expression and Disease. Nature Genetics, 37, 710, 2005

Benefits

• Ability to distinguish causality that are not discerned by state-of-the-arts methods (e.g. can infer causality in equivalence classes of Bayesian-network structures)
• No limitations on the nature of the data (can use continuous as oppose to discrete data)
• No limitations on the complexity and nature of the network (i.e. the graphs that describes the
• network can be of any type)
• Can work with static snapshot data and/or time-series observations
• Can generate future in-silico predictions on levels of the nodes in the network given perturbation

Applications

• Applicable in a broad range of industries where large datasets are analyzed to predict causal relationships (biomedical, financial, telecommunications, etc.)