It is my pleasure to recommend the paper by Raynal et al. [1] about using random forest for parameter inference. There are two reviews about the paper, one review written by Dennis Prangle and another review written by myself. Both reviews were positive and included comments that have been addressed in the current version of the preprint.

The paper nicely shows that modern machine learning approaches are useful for Approximate Bayesian Computation (ABC) and more generally for simulation-driven parameter inference in ecology and evolution.

The authors propose to consider the random forest approach, proposed by Meinshausen [2] to perform quantile regression. The numerical implementation of ABC with random forest, available in the abcrf package, is based on the RANGER R package that provides a fast implementation of random forest for high-dimensional data.

According to my reading of the manuscript, there are 3 main advantages when using random forest (RF) for parameter inference with ABC. The first advantage is that RF can handle many summary statistics and that dimension reduction is not needed when using RF.

The second advantage is very nicely displayed in Figure 5, which shows the main result of the paper. If correct, 95% posterior credibility intervals (C.I.) should contain 95% of the parameter values used in simulations. Figure 5 shows that posterior C.I. obtained with rejection are too large compared to other methods. By contrast, C.I. obtained with regression methods have been shrunken. However, the shrinkage can be excessive for the smallest tolerance rates, with coverage values that can be equal to 85% instead of the expected 95% value. The attractive property of RF is that C.I. have been shrunken but the coverage is of 100% resulting in a conservative decision about parameter values.

The last advantage is that no hyperparameter should be chosen. It is a parameter free approach, which is desirable because of the potential difficulty of choosing an appropriate acceptance rate.

The main drawback of the proposed approach concerns joint parameter inference. There are many settings where the joint parameter distribution is of interest and the proposed RF approach cannot handle that. In population genetics for example, estimation of the severity and of the duration of the bottleneck should be estimated jointly because of identifiability issues. The challenge of performing joint parameter inference with RF might constitute a useful research perspective.
 

References
 

[1] Raynal L, Marin J-M, Pudlo P, Ribatet M, Robert CP, Estoup A. 2017. ABC random forests for Bayesian parameter inference. arXiv 1605.05537v4, https://arxiv.org/pdf/1605.05537
[2] Meinshausen N. 2006. Quantile regression forests. Journal of Machine Learning Research 7: 983-999. http://www.jmlr.org/papers/v7/meinshausen06a.html

The tragic 2014-2016 Ebola outbreak that resulted in more than 28,000 cases and 11,000 deaths in West Africa [1] has been a surprise to the scientific community. Before 2013, the Ebola virus (EBOV) was known to produce recurrent outbreaks in remote villages near tropical rainforests in Central Africa, never exceeding a few hundred cases with very high virulence. Both EBOV’s ability to circulate for several months in large urban human populations and its important mutation rate suggest that EBOV’s virulence could evolve and to some extent adapt to human hosts [2]. Up to now, the high virulence of EBOV in humans was generally thought to be maladaptive, the virus being adapted to circulating in wild animal populations (e.g. fruit bats [3]). As a logical consequence, EBOV virulence could be expected to decrease during long epidemics in humans. The present paper by Sofonea et al. [4] challenges this view and explores how, given EBOV’s life cycle and known epidemiological parameters, virulence is expected to evolve in the human host during long epidemics. The main finding of the paper is that there is no chance that EBOV’s virulence decreases in the short and long terms. The main underlying mechanism is that EBOV is also transmitted by dead bodies, which limits the cost of virulence. In itself the idea that selection should select for higher virulence in diseases that are also transmitted after host death will sound intuitive for most evolutionary epidemiologists. The accomplishment of the paper is to make a very strong case that the parameter range where virulence could decrease is very small. The paper further provides scientifically grounded arguments in favor of the safe management of corpses. Safe burial of corpses is culturally difficult to impose. The present paper shows that in addition to instantaneously decreasing the spread of the virus, safe burial may limit virulence increase in the short term and favor of less virulent strains in the long term. Altogether these results make a timely and important contribution to the knowledge and understanding of EBOV.

References

[1] World Health Organization. 2016. WHO: Ebola situation report - 10 June 2016.

[2] Kupferschmidt K. 2014. Imagining Ebola’s next move. Science 346: 151–152. doi: 10.1126/science.346.6206.151

[3] Leroy EM, Kumulungui B, Pourrut X, Rouquet P, Hassanin A, Yaba P, Délicat A, Paweska, Gonzalez JP and Swanepoel R. 2005. Fruit bats as reservoirs of Ebola virus. Nature 438: 575–576. doi: 10.1038/438575a

[4] Sofonea MT, Aldakak L, Boullosa LFVV and Alizon S. 2017. Can Ebola Virus evolve to be less virulent in humans? bioRxiv 108589, ver. 3 of 19th May 2017; doi: 10.1101/108589