CoChat

A data-driven approach for predicting the impact of drugs on the human microbiome

Open access
Algavi, Yadid M., Borenstein, Elhanan · 2023 · Nature Communications

Published version of a preprint

The DOI matches its Crossref record; the references it cites are not checked yet.

Abstract

Many medications can negatively impact the bacteria residing in our gut, depleting beneficial species, and causing adverse effects. To guide personalized pharmaceutical treatment, a comprehensive understanding of the impact of various drugs on the gut microbiome is needed, yet, to date, experimentally challenging to obtain. Towards this end, we develop a data-driven approach, integrating information about the chemical properties of each drug and the genomic content of each microbe, to systematically predict drug-microbiome interactions. We show that this framework successfully predicts outcomes of in-vitro pairwise drug-microbe experiments, as well as drug-induced microbiome dysbiosis in both animal models and clinical trials. Applying this methodology, we systematically map a large array of interactions between pharmaceuticals and human gut bacteria and demonstrate that medications’ anti-microbial properties are tightly linked to their adverse effects. This computational framework has the potential to unlock the development of personalized medicine and microbiome-based therapeutic approaches, improving outcomes and minimizing side effects.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-023-39264-0

Cites

Statistics

Figures

Fig. 1 Machine-learning model prediction of in vitro drug–microbe interactions. A A scheme of our computational framework. B Receiver operating characteristic (ROC) curve for three learning settings: new drug–microbe interactions, new microbes, and new drugs. C Precision recall (PR) curve for the three learning settings as in panel B. D The upper panel displays the distribution of distances from the test compound to the nearest molecule in each cross-validation training set as a function of the TC distance cutoffs. The middle plot shows the distribution of the number of compounds in the training set. The lower panel illustrates the ROC AUC (yellow) and PR AUC (purple) for new drug prediction as a function of the TC distance cutoffs. E ROC AUC and PR AUC scores for the leave-one-microbe-out model as a function of phylogenetic distance (with Pearson correlation of −0.574, p = 5 × 10−4 for ROC AUC and −0.648, p = 8.23 × 10−6 for PR AUC, error bands show standard error). The color indicates the strain’s phyla. F Decrease in ROC AUC and PR AUC score when all strains for the same phylum are removed from the training set. The phyla Verrucomicrobia and Fusobacteria were discarded from this analysis as each contains only a single strain.
Fig. 1 Machine-learning model prediction of in vitro drug–microbe interactions. A A scheme of our computational framework. B Receiver operating characteristic (ROC) curve for three learning settings: new drug–microbe interactions, new microbes, and new drugs. C Precision recall (PR) curve for the three learning settings as in panel B. D The upper panel displays the distribution of distances from the test compound to the nearest molecule in each cross-validation training set as a function of the TC distance cutoffs. The middle plot shows the distribution of the number of compounds in the training set. The lower panel illustrates the ROC AUC (yellow) and PR AUC (purple) for new drug prediction as a function of the TC distance cutoffs. E ROC AUC and PR AUC scores for the leave-one-microbe-out model as a function of phylogenetic distance (with Pearson correlation of −0.574, p = 5 × 10−4 for ROC AUC and −0.648, p = 8.23 × 10−6 for PR AUC, error bands show standard error). The color indicates the strain’s phyla. F Decrease in ROC AUC and PR AUC score when all strains for the same phylum are removed from the training set. The phyla Verrucomicrobia and Fusobacteria were discarded from this analysis as each contains only a single strain.CC BY · PubMed Central
Fig. 2 The model’s drug and microbial feature importance scores. A The top 20 drug features with the highest feature importance score are colored according to the drug feature category. B The top 20 microbial features with the highest feature importance score are colored according to KEGG categories. All features included in these plots are statistically significant (p values calculated by permutation importance method, all values with FDR-corrected p < 0.05, exact values are provided in Supplementary Data 3).
Fig. 2 The model’s drug and microbial feature importance scores. A The top 20 drug features with the highest feature importance score are colored according to the drug feature category. B The top 20 microbial features with the highest feature importance score are colored according to KEGG categories. All features included in these plots are statistically significant (p values calculated by permutation importance method, all values with FDR-corrected p < 0.05, exact values are provided in Supplementary Data 3).CC BY · PubMed Central
Fig. 3 The landscape of drug impact on microbiome members. A Box plots describing the difference in impact scores between human target drugs and drugs targeting various microorganisms (pairwise comparison by Wilcoxon rank-sum test). A Kruskal–Wallis test further shows that the differences between ATC categories are statistically significant, p = 8.22 × 10−127, E2 = 0.138. A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively (Exact values are provided in Supplementary Data 5a). B The differences in impact scores between human-targeted drugs according to anatomical therapeutic chemical (ATC) classifications. The symbols above each label indicate the statistical significance in comparison with the mean impact score and the black points indicate the category means (pairwise comparison by Wilcoxon rank-sum test, exact values are provided in Supplementary Data 5b). A Kruskal–Wallis test further shows that the differences between ATC categories are statistically significant, p = 2.17 × 10−64, E2 = 0.102. C Phylogenetic tree for 409 microbiome members constructed based on their 16s rRNA gene. The outer circle denotes the drug sensitivity index, and the inner circle denotes the phylum. Taxa marked with green dots are those included in the in vitro screen (n = 39). ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.
Fig. 3 The landscape of drug impact on microbiome members. A Box plots describing the difference in impact scores between human target drugs and drugs targeting various microorganisms (pairwise comparison by Wilcoxon rank-sum test). A Kruskal–Wallis test further shows that the differences between ATC categories are statistically significant, p = 8.22 × 10−127, E2 = 0.138. A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively (Exact values are provided in Supplementary Data 5a). B The differences in impact scores between human-targeted drugs according to anatomical therapeutic chemical (ATC) classifications. The symbols above each label indicate the statistical significance in comparison with the mean impact score and the black points indicate the category means (pairwise comparison by Wilcoxon rank-sum test, exact values are provided in Supplementary Data 5b). A Kruskal–Wallis test further shows that the differences between ATC categories are statistically significant, p = 2.17 × 10−64, E2 = 0.102. C Phylogenetic tree for 409 microbiome members constructed based on their 16s rRNA gene. The outer circle denotes the drug sensitivity index, and the inner circle denotes the phylum. Taxa marked with green dots are those included in the in vitro screen (n = 39). ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.CC BY · PubMed Central
Fig. 4 Prediction of in vivo drug-induced dysbiosis. Violin plots illustrating the difference in predicted impact scores between taxa with significantly increased vs. decreased abundance following drug administration of A Omeprazole (human clinical trial, n = 153 taxa), B Paclitaxel (mouse model, n = 120 taxa), and C Methotrexate (rat model, n = 149 taxa). Statistics were calculated by a two-sided Wilcoxon rank-sum test. A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively.
Fig. 4 Prediction of in vivo drug-induced dysbiosis. Violin plots illustrating the difference in predicted impact scores between taxa with significantly increased vs. decreased abundance following drug administration of A Omeprazole (human clinical trial, n = 153 taxa), B Paclitaxel (mouse model, n = 120 taxa), and C Methotrexate (rat model, n = 149 taxa). Statistics were calculated by a two-sided Wilcoxon rank-sum test. A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively.CC BY · PubMed Central
Fig. 5 Comparison between model predictions and drug-induced dysbiosis in the Lifelines cohort. A Correlation between the average predicted impact per microbial genus and the taxa-averaged observed negative impact in the Lifelines cohort. (n = 84 taxa; error bands show standard error) B Correlation between the average predicted impact per drug group and drug-averaged observed impact in the Lifelines cohort. (n = 40 drugs; error bands show standard error) C, D Differences in predicted impact scores between taxa with positive vs negative association with two common drug groups, proton-pump inhibitors (C, Two-sided Wilcoxon rank-sum test, effect size 0.425, n = 66 taxa) and osmotic laxatives (D, Two-sided Wilcoxon rank-sum test, effect size = 0.466, n = 45 taxa). A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively.
Fig. 5 Comparison between model predictions and drug-induced dysbiosis in the Lifelines cohort. A Correlation between the average predicted impact per microbial genus and the taxa-averaged observed negative impact in the Lifelines cohort. (n = 84 taxa; error bands show standard error) B Correlation between the average predicted impact per drug group and drug-averaged observed impact in the Lifelines cohort. (n = 40 drugs; error bands show standard error) C, D Differences in predicted impact scores between taxa with positive vs negative association with two common drug groups, proton-pump inhibitors (C, Two-sided Wilcoxon rank-sum test, effect size 0.425, n = 66 taxa) and osmotic laxatives (D, Two-sided Wilcoxon rank-sum test, effect size = 0.466, n = 45 taxa). A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively.CC BY · PubMed Central
Fig. 6 Association of anti-commensal activity and adverse drug reactions. Box plots illustrate the difference in predicted impact scores between drugs with A high vs. low frequency of gastrointestinal drug adverse effects and B high vs. low frequency of infectious drug adverse effects. C Difference in diarrhea frequency between drugs with impact specificity to Firmicutes and Bacteroidetes, as well as drugs without specificity to any phyla (n = 771 drugs, two-sided Wilcoxon rank-sum test). A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively.
Fig. 6 Association of anti-commensal activity and adverse drug reactions. Box plots illustrate the difference in predicted impact scores between drugs with A high vs. low frequency of gastrointestinal drug adverse effects and B high vs. low frequency of infectious drug adverse effects. C Difference in diarrhea frequency between drugs with impact specificity to Firmicutes and Bacteroidetes, as well as drugs without specificity to any phyla (n = 771 drugs, two-sided Wilcoxon rank-sum test). A line across the box indicates the median. The whiskers are lines extending from Q1 and Q3 to endpoints that are defined as the most extreme data points within Q1 − 1.5 × IQR and Q3 + 1.5 × IQR, respectively.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

The citation lists are queued

They are fetched later and appear here then.

References 80 · Cited by 54

Metadata

Authors
Algavi, Y. M., & Borenstein, E.
Affiliations
  • Tel Aviv University (IL)
  • Santa Fe Institute (US)
Corresponding author
Borenstein, Elhanan
Venue
Nature Communications
Published
2023-06-17
Volume
14
Issue
1
Pages
3614-3614
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4381051893
Cited by
54
References
80
Citations by year
2026: 22 · 2025: 20 · 2024: 12
FWCI
4.79
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • U.S. Department of Health & Human Services | National Institutes of Health (U19AG057377)
  • Israel Science Foundation (2435/19)
  • Faculty Fellow of the Edmond J. Safra Center for Bioinformatics at Tel Aviv University.
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gut microbiota and health
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Health Sciences › Dentistry › Periodontics › Oral microbiology and periodontitis research
Keywords
drug-microbiome interactions, gut microbiota, adverse drug reactions, personalized medicine, microbiota-based therapy
MeSH terms
Gastrointestinal Microbiome, Animals, Humans, Genomics, Microbiota, Drug-Related Side Effects and Adverse Reactions, Dysbiosis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A data-driven approach for predicting the impact of drugs on the human microbiome. https://app.cochat.ai/papers/10.1038/s41467-023-39264-0
The paper
Algavi, Y. M., & Borenstein, E. (2023). A data-driven approach for predicting the impact of drugs on the human microbiome. Nature Communications. https://doi.org/10.1038/s41467-023-39264-0

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.