Office of Technology Transfer – University of Michigan

A Dynamic Bayesian Environment for Learning and Predicting Trends in Biomedical Data

Technology #3792

Bayesian network-based computer algorithm for life science applications

Based on Bayesian network (BN) modeling, this software combines microarray gene expression data with pathway models to facilitate discovery of new factors that are involved in biochemical pathways. This can help identify new therapeutic targets for drug discovery efforts. Additionally, a computer algorithm that models temporal biomedical data using dynamic Bayesian networks (DBN) can be used to discover and validate biomarkers for patient response to single or combination therapies, and monitor efficacy of a therapeutic intervention in real-time. These predictions can be used in clinical trials for patient selection and efficacy assessment, and provide time and cost-effective solutions for clinical evaluation of new therapies.

Drug development process is a time-consuming multi-step process with high attrition rates due to poor pharmacokinetics, lack of efficacy, toxicity, and adverse effects. More than 90% of all new oncology drugs that enter clinical trials do not obtain marketing approval. With clinical trials increasing in complexity, cost and difficulty to execute, drug discovery organizations are under growing pressure to reduce costs and safely improve R&D efficiencies. Bioinformatics and data analytics software tools that can help streamline drug development and improve outcomes can provide valuable cost-saving information.

In silico methods to improve R&D efficiency

Although correlation-based methods are widely used for analyzing large datasets such as gene expressing profiling, they lack the ability to establish causal relationships between different network factors. In contrast, Bayesian network (BN) analysis is able to establish causality, or directionality, between nodes. Additionally, Bayesian networks can be expanded to evaluate if additional factors can improve specific networks, and thus enable identification of novel pathway elements.

Researchers at the University of Michigan developed a BN-based software capable of identifying new pathway elements. To validate this BN expansion algorithm, gene expression data from perturbed B cells from AfCS database was used to construct a core BN using B cell receptor (BCR) pathway as found in KEGG database. Subsequently, this core BN was tested with genes not present in the current KEGG BCR pathway, and a new BCR pathway protein Rexo2 was identified as its addition to the core BN improved its probability score. Biochemical data on Rexo2 confirms its role in the BCR pathway, and validates this approach for new pathway protein identification.

Algorithm based on dynamic Bayesian networks (DBN) was developed that can identify biomarkers to predict disease progress, efficacy of new therapeutic interventions, and suggest drug combinations with maximum benefit for specific patients. In this algorithm DBNs are constructed based on biological data acquired at various time points, and subsequently assigned probability scores. Simulations are run to reflect therapeutic interventions, and BNs are rescored such that they incorporate results of these simulations as if they were experimental data points. Based on these changes, therapeutic interventions are evaluated for their ability to either maximize differences (maximum efficacy) or minimize differences (minimize side effects).

These algorithms can be used in a wide variety of applications in the life sciences and healthcare industries. They can be used for target identification, target validation by modeling and simulation of specific interventions, biomarker identification, efficacy and toxicity predictions, patient selection in clinical trials and by clinicians, identification of most efficacious drug combinations, in adaptive clinical trial design, and to provide rapid feedback on drug efficacy during clinical trials.

Applications and Advantages


  • Bioinformatics software that integrates gene expression profiling and biochemical pathway analysis
  • Identify new drug targets
  • Identify biomarkers to predict disease progress and drug efficacy
  • Predict efficacy of multi-drug combinations
  • Software for design of adaptive clinical trials


  • Decrease cost and time of clinical trials
  • Improve R&D efficiency