Peer review process

Not revised: This Reviewed Preprint includes the authors’ original preprint (without revision), an eLife assessment, public reviews, and a response from the authors (if available).

Read more about eLife’s peer review process.

Editors

  • Reviewing Editor
    Lukas Folkman
  • Senior Editor
    Wendy Garrett
    Harvard T.H. Chan School of Public Health, Boston, United States of America

Reviewer #1 (Public Review):

Summary:

De Waele et al. reported a dual-branch neural network model for predicting antibiotic resistance profiles using matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry data. Neural networks were trained on the recently available DRIAMS database of MALDI-TOF mass spectrometry data and their associated antibiotic susceptibility profiles. The authors used a dual branch neural network approach to simultaneously represent information about mass spectra and antibiotics for a wide range of species and antibiotic combinations. The authors showed consistent performance of their strategy to predict antibiotic susceptibility for different spectrums and antibiotic representations (i.e., embedders). Remarkably, the authors showed how small datasets collected at one location can improve the performance of a model trained with limited data collected at a second location. Despite these promising results, there are several analyses that the authors could incorporate to offer additional support to some of their claims (see weaknesses). In particular, this work would benefit from a more comprehensive comparison of the author's single recommender model vs an ensemble of specialist models, and the inclusion of 1-2 examples that showcase how their model could be translated into the clinic.

Strengths:

• A single AMR recommender system could potentially facilitate the adoption of MALDI-TOF-based antibiotic susceptibility profiling into clinical practices by reducing the number of models to be considered, and the efforts that may be required to periodically update them.

• Authors tested multiple combinations of embedders for the mass spectra and antibiotics while using different metrics to evaluate the performance of the resulting models. Models trained using different spectrum embedder-antibiotic embedder combinations had remarkably good performance for all tested metrics. The average ROC AUC scores for global and spectrum-level evaluations were above 0.9. Average ROC AUC scores for antibiotic-level evaluations were greater than 0.75.

• Authors showed that data collected in one location can be leveraged to improve the performance of models generated using a smaller number of samples collected at a different location. This result may encourage researchers to optimize data integration to reduce the burden of data generation for institutions interested in testing this method.

Weaknesses:

• Although ROC AUC is a widely used metric. Other metrics such as precision, recall, sensitivity, and specificity are not reported in this work. The last two metrics would help readers understand the model's potential implications in the context of clinical research.

• The authors did not hypothesize or describe in any way what an acceptable performance of their recommender system should be in order to be adopted by clinicians.

• Related to the previous comment, this work would strongly benefit from the inclusion of 1-2 real-life applications of their method that could showcase the benefits of their strategy for designing antibiotic treatment in a clinical setting.

• The authors do not offer information about the model features associated with resistance. This information may offer insights about mechanisms of antimicrobial resistance and how conserved they are across species.

• Comparison of AUC values across models lacks information regarding statistical significance. Without this information it is hard for a reader to figure out which differences are marginal and which ones are meaningful (for example, it is unclear if a difference in average AUC of 0.02 is significant). This applied to Figure 2, Figure 3, and Table 2 (and the associated supplementary figures).

• One key claim of this work was that their single recommender system outperformed specialist (single species-antibiotic) models. However, in its current status, it is not possible to determine that in fact that is the case (see comment above). Moreover, comparisons to species-level models (that combine all data and antibiotic susceptibility profiles for a given species) would help to illustrate the putative advantages of the dual branch neural network model over species-based models. This analysis will also inform the species (and perhaps datasets) for which specialist models would be useful to consider.

• Taking into account that the clustering of spectra embeddings seemed to be species-driven (Figure 4), one may hypothesize that there is limited transfer of information between species, and therefore the neural network model may be working as an ensemble of species models. Thus, this work would deeply benefit from a comparison between the authors' general model and an ensemble model in which the species is first identified and then the relevant species recommender is applied. If authors had identified cases to illustrate how data from one species positively influence the results for another species, they should include some of those examples.

Reviewer #2 (Public Review):

The authors frame the MS-spectrum-based prediction of antimicrobial resistance prediction as a drug recommendation task. Weis et al introduced the dataset this model is tested on and benchmark models which take as input a single species and are trained to predict resistance to a single drug. Instead here, a pair of drug and spectrum are fed to 2 neural network models to predict a resistance probability. In this manner, knowledge from different drugs and species can be shared through the model parameters. Three questions are asked: 1. what is the best way to encode the drugs? 2. does the dual NN outperform the single-spectrum drug?

Overall the paper is well-written and structured. It presents a novel framework for a relevant problem. The work would benefit from more work on evaluation.

  1. Howard Hughes Medical Institute
  2. Wellcome Trust
  3. Max-Planck-Gesellschaft
  4. Knut and Alice Wallenberg Foundation