Bayesian networks elucidate complex genomic landscapes in cancer

Area of Science:

• Artificial Intelligence
• Computational Biology
• Genomics

Background:

• Bayesian networks (BNs) are explainable AI models for structured probability spaces.
• BNs can be constructed from observed data to explore complex relationships in biological settings.

Purpose of the Study:

• To elucidate relationships between driver events in large cancer genomic datasets using BNs.
• To provide a methodology tailored for biologists and clinicians producing such datasets.

Main Methods:

• Utilizing an optimal BN learning algorithm with well-established likelihood functions.
• Employing two intuitive tuning parameters for ease of use.
• Incorporating heatmaps for network families and visualizing pairwise co-occurrence statistics.

Main Results:

• Demonstrated enhanced pairwise testing capabilities.
• Applied the methodology to 5 cancer datasets, revealing complex genomic landscapes.
• Identified a 4-way mutual exclusivity in myeloma and a 3-way mutual exclusivity in myeloproliferative neoplasms.

Conclusions:

• The proposed BN methodology is valuable for discussing driver event relationships in large genomic cohorts.
• This approach can play a central role in the study of large genomic cancer datasets.