Research tool
Ecological dysbiosis score
The score ranks a faecal 16S rRNA sample against a reference cohort of 10,276 samples, including controls and cases from 13 diseases. It indicates how far the community deviates in the direction the model associates with disease. It does not identify a specific disease, estimate a probability, or indicate whether an individual is ill.
For research use only. This tool is not a medical device and is not intended for diagnosis. For health concerns, please consult a clinician.
1. Select input data
A tab-separated table with SILVA 138.2 OTU identifiers (97% identity) as rows, in the form accession.start.stop, and one column per sample. This format can be exported from QIIME 2 and DADA2.
Example file: otu_table_example.tsv (two samples: a CRC control and a CRC case).
Or load a sample from the reference cohort:
Amplicon sequences and their counts. Sequences are mapped to the reference OTUs on the server; this is the only step not performed in the browser, and no data are stored.
Example files: rep_seqs_example.fasta and asv_table_example.tsv (the same two samples as 300-bp V4 amplicon sequence variants).
2. Run
3. Result
Limitations
Please read before interpreting the score
The model was trained only on human faecal 16S rRNA data. The score is not valid for soil, water, oral, skin or vaginal communities, or for data other than 16S rRNA amplicon sequencing.
The reference cohort covers 13 diseases, and the model was trained on all of them. For a disease outside those 13 the expected performance is AUC 0.64, measured by leave-one-disease-out validation. The pooled leave-one-disease-out model reported in the paper reaches 0.67 on the same cohort, so the deployed ensemble scores an unseen disease at close to the performance the paper reports; both figures are held-out estimates. Samples from conditions not represented in the cohort are likely to receive scores closer to the middle of the distribution.
The score is derived from a single sample using a single model. Within-individual variation in gut communities over a month can exceed the differences the score is designed to detect. A single sample is therefore not sufficient to draw conclusions about an individual.