Episode 17: Bayesian magic in practice

The microbinfie podcast explores the practical applications of Bayesian methods in bioinformatics and phylogenetics, providing insights into computational techniques for understanding evolutionary relationships.

Join Dr. Conor Meehan, Dr. Leo Martins, and Dr. Nabil-Fareed Alikhan as they delve deeper into the fascinating realm of Bayesian methodologies. This part of the series continues the exploration of the Bayesian approach in various fields, shedding light on its applications and intricacies.

Guests

• Dr. Conor Meehan
• Dr. Leo Martins

Some notes

• Bayesian Models:

  • Not a philosophy or religion, but a class of models based on conditional probabilities.
  • Integrate various data sources for phylogenetic analysis.

• Key Concept:

  • Bayes' Theorem: The probability of a model given the data equals the probability of the data given the model multiplied by the prior probability of the model.

• Priors and Posteriors:

  • Prior: Initial belief about parameters before observing data.
  • Posterior: Updated belief after incorporating data.

Getting Started with Bayesian Inference

• Tools:

  • Rev: Recommended for beginners; simple tutorials available.
  • STAN: Known for graphical models and flowcharts to describe interactions.

• Resources:

  • Taming the Beast: An online course with tutorials for various skill levels.

Building Bayesian Methods into Existing Work

• Collaboration: Encouraged to consult statisticians for guidance and validation.
• Simulation Techniques:
  • Simulated Annealing: Optimization technique for Bayesian models.
  • Approximate Bayesian Computing (ABC): Simulates data to approximate posterior distributions.

Interpreting Results

• Convergence Checking: Essential to ensure analysis is complete.

  • Tools:
    • Tracer: Commonly used for visualizing convergence.
    • RWTY: Computes distances between trees and checks convergence.

• Estimates:

  • Most common output is the Maximum A Posteriori (MAP) tree.
  • Programs like Big Tree help in visualizing uncertainties.

Popular Bayesian Software in Phylogenetics

• BEAST:

  • Well-known for creating time trees with a user-friendly interface.
  • Comes with tools for post-processing and analyzing convergence.

• MrBayes and RevBayes:

  • MrBayes is no longer supported; RevBayes is the recommended alternative.

• Comparative Tools:

  • LSD: For creating time trees from maximum likelihood trees.

Models in Bayesian Inference

• Bayesian models are used extensively for phylogenetic analysis, but they require careful consideration of priors and data inputs.
• Understanding the differences between model outputs in Bayesian vs. maximum likelihood frameworks is crucial.

Bayesian inference in microbial bioinformatics allows for complex modeling and integration of various data sources, offering a robust framework for understanding evolutionary relationships among pathogens.