A droplet-to-data workflow for antimicrobial screening of complex matrices
Screening plant extracts and other complex natural matrices for antibacterial activity is complicated by optical interference and population averaging in conventional bulk assays. Droplet-based assays mitigate interference while enabling high-throughput assessment of bacterial subpopulations. However, analyzing the resulting image data involves difficult image processing, classification, and interpretation. Here, we present a democratized workflow for antimicrobial screening in complex matrices that combines a polydisperse droplet-based assay with EasyFlow 2.0 (EF2), our updated droplet image analysis platform. EF2 offers stable containerized deployment, expanded polydisperse droplet support, automated thresholding and MIC estimation, LLM-assisted data exploration, and import of other software outputs such as CellProfiler (CP) measurements and CellProfiler Analyst (CPA) classification models. As proof of concept, we applied this workflow to assess the antibacterial and aggregation-modulating effects of chemically distinct extracts of Taraxacum officinale, Rhodiola rosea, and Lonicera caerulea against Escherichia coli. Extracts were chemically characterized by HPLC-MS and evaluated using both standard 96-well plate microdilution and droplet-based assays. In total, 27,059 droplets were segmented in CP, classified into four growth phenotypes using a CPA random forest model, and analyzed in EF2. The droplet-based assay reached minimal inhibitory concentration (MIC) at lower extract concentrations than 96-well plate microdilution and revealed extract-specific aggregation effects: L. caerulea reduced E. coli aggregation, whereas T. officinale did not. R. rosea reduced growth in the 96-well plate microdilution assay, but high concentrations caused droplet gelation that prevented droplet-based analysis, showing that sample components must be checked for compatibility with droplet formation. These results illustrate that the EF2-centered workflow supports rapid, automated MIC estimation through unsupervised thresholding when labels are unavailable, and four-class aggregation phenotyping when a CPA model is supplied. This workflow offers flexible, scalable screening, potentially extendable beyond plant extracts to other complex sample matrices.