Understanding the diversity of soil organisms is complicated by both scale and substrate. Every footprint we leave in the soil covers hundreds to millions of organisms yet we cannot see them without extremely laborious extraction and microsopy endeavors. Studying them is also challenging. Keeping them alive so that we can understand their lifecycles and ecological roles ranges from difficult to impossible. Functional and taxonomic identification of soil organisms, while possible, is also challenging. Here we present the Smart Soil Organism Detector, an instrument and machine learning pipeline that combines high-resolution imaging, multi-spectral sensing, large-bore flow cytometry, and machine learning to extract, isolate, count, identify, and separate soil organisms in a high-throughput, high-resolution, non-destructive, and reproducible manner. This system is not only capable of separating alive nematodes, dead nematodes, and nematode cuticles from soil with 100% out-of-sample accuracy, but also capable of identifying nematode strains (sub-species) with 95.5% out-of-sample accuracy and 99.4% specificity. Soil micro-arthropods were identified to class with 96.1% out-of-sample accuracy. Broadly applicable across soil taxa, the Smart SOD system is a tool for understanding global soil biodiversity.
Keywords: Deep learning; Flow cytometry; High throughput sensing; Micro-arthropods; Nematodes; Soil biodiversity.
Copyright © 2022 Elsevier B.V. All rights reserved.