Deep learning and likelihood approaches for viral phylogeography converge on the same answers whether the inference model is right or wrong

Abstract

Analysis of phylogenetic trees has become an essential tool in epidemiology. Likelihood-based methods fit models to phylogenies to draw inferences about the phylodynamics and history of viral transmission. However, these methods are computationally expensive, which limits the complexity and realism of phylodynamic models and makes them ill-suited for informing policy decisions in real-time during rapidly developing outbreaks. Likelihood-free methods using deep learning are pushing the boundaries of inference beyond these constraints. In this paper, we extend, compare and contrast a recently developed deep learning method for likelihood-free inference from trees. We trained multiple deep neural networks using phylogenies from simulated outbreaks that spread among five locations and found they achieve similar levels of accuracy to Bayesian inference under the true simulation model. We compared robustness to model misspecification of a trained neural network to that of a Bayesian method. We found that both models had comparable performance, converging on similar biases. We also trained and tested a neural network against phylogeographic data from a recent study of the SARS-Cov-2 pandemic in Europe and obtained similar estimates of epidemiological parameters and the location of the common ancestor in Europe. Along with being as accurate and robust as likelihood-based methods, our trained neural networks are on average over 3 orders of magnitude faster. Our results support the notion that neural networks can be trained with simulated data to accurately mimic the good and bad statistical properties of the likelihood functions of generative phylogenetic models.

Trait-mediated speciation and human-driven extinctions in proboscideans revealed by unsupervised Bayesian neural networks

“…Neural networks have been recently used to predict speciation and extinction rates from fossil or phylogenetic data under simple age- or trait-dependent models ( 67 – 69 ). These methods were applied within a supervised learning framework, in which (i) labeled training datasets (fossil occurrences or phylogenies) are simulated under known speciation and extinction rates; (ii) neural networks are trained to predict the generating rates from features describing the simulated data; and (iii) the trained models are then applied to the unlabeled empirical dataset, where the rates are unknown.…”

Recent evolutionary origin and localized diversity hotspots of mammalian coronaviruses

“…We have shown by simulation that contrasting scenarios, such as codiversification, or diversification by preferential host switching after originating within or outside bats, leave distinct signatures in the data. This suggests that deep neural networks trained on simulated data, which have already shown their performance in phylodynamics (47,48,49,50, could also be useful for the analysis of cophylogenetic data.…”

PhyloJunction: a computational framework for simulating, developing, and teaching evolutionary models

“…We introduced PhyloJunction, an open-source package for simulating state-dependent speciation and extinction (SSE) processes, a large family of diversification models that has found success across a range of scientific domains [ 13 , 27 , 75 ]. Most implementations of SSE models have prioritized inference and efficiency over simulation and generality; the latter is the relatively vacant niche PhyloJunction was designed to fill.…”

“…Alternatively, it should be straightforward to integrate PhyloJunction’s functionalities for summarizing data together with Python machine learning libraries. There is increasing evidence [ 49 ] backing machinelearning methods as viable alternatives to frequentist and Bayesian evolutionary inference, especially when the latter is very onerous or impossible [ 75 ].…”