PhyloDeep

PhyloDeep applies deep learning to phylogenetic trees to perform likelihood-free phylodynamic inference and estimate epidemiological parameters.

Key Features:

• Likelihood-Free Simulation-Based Approach: PhyloDeep employs a likelihood-free, simulation-based methodology that circumvents explicit likelihood calculations used in maximum-likelihood and Bayesian approaches.
• Deep Learning Integration: The software integrates deep learning models that process either a comprehensive set of summary statistics derived from phylogenies or a complete, compact representation of trees.
• Scalability and Speed: It efficiently handles very large phylogenetic datasets, offering rapid processing and high accuracy relative to existing state-of-the-art methods.
• Model Selection and Parameter Estimation: PhyloDeep supports robust model selection and accurate estimation of epidemiological parameters from genetic data.
• Analysis of Complex Transmission Phenomena: The approach enables assessment of complex phenomena such as superspreading events.

Scientific Applications:

• Phylodynamic Model Selection: PhyloDeep facilitates selection among competing phylodynamic models using simulation-based inference.
• Epidemiological Parameter Estimation: It enables estimation of epidemiological parameters from extensive genetic datasets to inform outbreak dynamics.
• Infectious Disease Research: The method is applicable to studies of outbreak dynamics and transmission processes in infectious disease research.
• HIV Case Study: PhyloDeep has been applied to an HIV dataset from men having sex with men in Zurich to assess epidemiological dynamics including superspreading.

Methodology:

PhyloDeep uses deep learning models trained on simulated datasets representing evolutionary and epidemiological scenarios in a likelihood-free, simulation-based framework and processes either summary statistics or compact tree representations to infer model choice and epidemiological parameters.