Unsupervised Analysis of Transcriptomics in Bacterial Sepsis Across Multiple Datasets Reveals Three Robust Clusters

Abstract

Objectives: To find and validate generalizable sepsis subtypes using data-driven clustering.

Design: We used advanced informatics techniques to pool data from 14 bacterial sepsis transcriptomic datasets from eight different countries (n = 700).

Setting: Retrospective analysis.

Subjects: Persons admitted to the hospital with bacterial sepsis.

Interventions: None.

Measurements and main results: A unified clustering analysis across 14 discovery datasets revealed three subtypes, which, based on functional analysis, we termed "Inflammopathic, Adaptive, and Coagulopathic." We then validated these subtypes in nine independent datasets from five different countries (n = 600). In both discovery and validation data, the Adaptive subtype is associated with a lower clinical severity and lower mortality rate, and the Coagulopathic subtype is associated with higher mortality and clinical coagulopathy. Further, these clusters are statistically associated with clusters derived by others in independent single sepsis cohorts.

Conclusions: The three sepsis subtypes may represent a unifying framework for understanding the molecular heterogeneity of the sepsis syndrome. Further study could potentially enable a precision medicine approach of matching novel immunomodulatory therapies with septic patients most likely to benefit.

Figure 1

Overall study schematic.

Figure 2

The first two principal components (PCs) of the discovery clustering results (both with (A) and without (B) the 16% of samples that went unclustered in the final analysis, in gold) using all 8,946 genes present in the COCONUT conormalized data. Here we show that the cluster assignments that we recovered in an unsupervised manner are clearly separated in high-dimensional space, as demonstrated by the first two principal components.

Figure 3

(A) Correlations of average 500-gene expression vectors between clusters assigned in the discovery and validation datasets; correlation coefficient is shown by color (legend at figure right). Notably, samples from Inflammopathic clusters are positively correlated with Inflammopathic samples from other datasets, and negatively correlated with Adaptive samples from other datasets (and vice-versa). The Coagulopathic clusters show less cohesion but are positively correlated with one another. (B) Heatmap of Gene Ontology (GO) codes found to be overrepresented in the different clusters, colored by significance levels. In both (A) and (B), the pooled ‘Core’ discovery datasets are represented by a single column for each cluster, while each cluster in each validation dataset is represented by a separate column. Both sub-figures show a block structure indicative of molecular similarity across datasets between clusters of the same type.