Raman Single-Cell spectroscopy (SCRS) is among the most promising technologies to characterize microbial ecology, which reveals single-cell level insights of microbial phenotypes in complementary to main-stream genotypic technologies. SCRS fingerprints chemical composition of un-cultured cells by measuring the spectra of photons that are scattered to a different frequency than the monochromatic incident laser. A major challenge in the application of single-cell technologies, such as SCRS, is the determination of sufficient sampling size in the balance between statistical power and the limitation on time, labor and resources. Current SCRS-based microbial ecology studies often assume a randomly chosen sample size, or determine the size based on prior knowledge without the ability to deduce true organism diversity in environmental samples. In this study, we proposed, tested and validated a computational sampling size planning protocol, taking advantage of a novel metric, <i>kernel divergence</i>, which compares dataset-wise distributional differences. The proposed method relies on no prior knowledge or extra human intervention. In addition, respective tests verified that its algorithm can appropriately handle datasets with both linear and non-linear relationships. These advantages promise the proposed method to be a standardized approach for sampling size optimization, enabling more comparable and reproducible experiments and analysis. We validated this model with 8 SCRS phenotyping datasets each sampled from a different enhanced biological phosphorus removal (EBPR) microbial community located across North America. Results suggested that 40-60 spectra were likely sufficient for EBPR-like communities. Further experiments revealed that these communities had rather similar sample distributions, indicating that they share more common and intrinsic phenotypic characteristics despite the geological, operational and configuration differences of these wastewater treatment facilities. Finally, though not tested, we expect that this technique can be easily generalized to other environmental applications and datasets as long as they contain only continuous features.

Machine learning kernel divergence-based <i>de novo</i> sampling size optimization for environmental applications: Cases study of single-cell Raman spectroscopic phenotyping of EBPR microbial communities

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.

Machine learning kernel divergence-based de novo sampling size optimization for environmental applications: Cases study of single-cell Raman spectroscopic phenotyping of EBPR microbial communities