Preprints reviewed by eLife: Applying deep learning to the microbiome

Read what peer reviewers thought about a recent preprint on the use of deep learning to design microbial communities with specific metabolite profiles.
Inside eLife
  • Views 252
  • Annotations

eLife only reviews papers that have been posted as preprints and, when the authors agree, we post some of the outcomes of the peer review process alongside the preprint. Authors can also post a response to this material if they wish.

During the review process our referees write ‘public reviews’ that discuss the strengths and limitations of the preprint, and our editors synthesize these into a short ‘evaluation summary’ that encapsulates their views on the paper. Reviewers also produce detailed feedback for the authors, including requests for revisions and suggestions for improvement; this feedback is not posted alongside the preprint.

This is the third in a series of short articles highlighting preprints that had been reviewed in this way. Links to the previous articles are given at the end of this article.

Preprint title: Deep learning enables design of multifunctional synthetic human gut microbiome dynamics

In this bioRxiv preprint the authors use an approach from deep learning to model the dynamics of communities of microbes. This is challenging because of the complexity of the interactions between the microbes and the metabolites they produce. The authors showed that an approach from deep learning called the long short-term memory (LSTM) framework could be used to design communities of microbes with specific health-relevant profiles of metabolites, and that this approach was able to outperform a number of existing approaches.

Here is what the evaluation summary for the preprint said:

The ultimate goal of this work is to apply machine learning to learn from experimental data on temporal dynamics and functions of microbial communities to predict their future behavior and design new communities with desired functions. Using a significant amount of experimental data, the authors suggest that their method outperforms a commonly used approach. Overall, the work is potentially of broad interest to those working on microbiome prediction and design.

The three reviewers went into the strengths and weaknesses of the work in more detail in their public reviews. Reviewer #1 summarized the strengths as follows:

This is an interesting study demonstrating the application of deep learning to model microbiome dynamics of the human gut community, improving on existing approaches (for example regarding scalability). Furthermore, the model is able to better predict microbe-microbe and microbe-metabolite interactions as compared to classical approaches like ODEs or regression. The authors show that their LSTM-based model is able to successfully predict the abundance not only at the final time step but also at intermediate time steps. In general, the authors did a good job in demonstrating the strengths of their proposed approach. The major findings were carefully interpreted and challenged through multiple tests (explainability and sensitivity analysis).

Reviewer #3 identified six “points of weakness” in their public review:

- It is unclear why an LSTM would be a good model for the microbiome
- It is unclear what aspects of the dynamics are long-term, and whether the experiments capture this long-term effect
- Discussions around the LSTM model and some ML and dynamical systems concepts are inaccurate (LSTM with one hidden unit is not really a "deep" model, gLV models are linear in the parameters and thus the parameters are trivial to solve for give the microbial abundances)
- Not enough detail is given regarding the LSTM model or the composite model to understand them
- part of the composite model is in Matlab and could not be tested
- authors claim that their model is interpretable, but it is no more interpretable than any differentiable model that can use gradients to open the lid after training

Reviewer #3 also commented on the likely impact on the field:

The tight coupling of new experiments with computational methods is important. All too often a tool is made but only shown to work on data not tailored to the tool. Here, both are designed together.

The evaluation summary and public reviews were posted on March 31, 2022.

Links to previous articles:

Peer Review: Highlighting preprints reviewed by eLife during February

Preprints reviewed by eLife: A "tour-de-force" study of neuronal maturation