Machine learning processing of microalgae flow cytometry readings: illustrated with Chlorella vulgaris viability assays

Abstract

A flow cytometry viability assay protocol is proposed and applied to model microalgae Chlorella vulgaris. The protocol relies on concomitant dual staining of the cells (Fluorescein DiAcetate (FDA), Propidium Iodide (PI)) and machine learning processing of the results. Protocol development highlighted that working at 4°C allows to preserve the stained sample for 15 minutes before analysis. Furthermore the inclusion of an extracellular FDA washing step in the protocol improves the signal to noise ratio, allowing better detection of active cells. Once established, this protocol was validated against 7 test cases (controlled mixtures of active and non-viable cells). Its performances on the test cases is good:-0.19 %abs deviation on active cell quantification (processed by humans). Furthermore, a machine learning workflow, based on DBSCAN algorithm, was introduced. After a calibration procedure, the algorithm provided very satisfactorily results with-0.10 %abs deviation compared to human processing. This approach permitted to automate and speed up (15 folds) cytometry readings processing. Finally, the proposed workflow was used to assess Chlorella vulgaris cryostorage procedure efficiency. The impact of freezing protocol on cells viability was first investigated over 48 hours storage (-20°C). Then the most promising procedure (pelleted,-20°C) was tested over one month. The observed trends and values cell in viability loss correlate well with literature. This shows that flow cytometry is a valid tool to assess for microalgae cryopreservation protocol efficiency.