CyanoflowSingle-cell microalgae research
A single-cell discovery concept for microalgae and cyanobacteria. Compartmentalize cells in a microfluidic chip, image them over time, extract measurable features, and let a model surface the individuals worth investigating.
In development: imagery and interactive consoles are illustrative, not experimental results. The point is resolution: instead of measuring the average of a whole culture, Cyanoflow would follow individual cells and turn that variety into searchable data.
- Per-cell records
- 100k target
- Observation
- Time-series
- Output
- Ranked candidates
Illustrative simulation — not instrument data
Single-cell methods: evidence before candidate claims
Read the microfluidic microalgae analysis guide for droplets versus microwells, time-series tracking and independent phenotype validation. The research methodology hub separates literature evidence from our proposed architecture; the glossary defines the terms.
Behrendt et al. (2020), PhenoChip, is a primary literature example of single-cell photophysiology. It is another research group's experiment, not validation of Cyanoflow.
From a drop of culture to a ranked list of individual cells
Each stage is a separable engineering problem. Stages 06 and 08 are shown live further down the page.
Sample
A prepared microalgal sample enters the system. Real environmental material contains many different cells and particles, so the first stage exists to make the input predictable enough for controlled analysis: filtration, dilution to a target cell density, and a defined suspension medium.
- Approved, well-characterized strain
- Filtered / diluted to target density
- Buffer + surfactant compatibility checked
- No casual culturing of unknown environmental microbes
Single-cell compartmentalization
The sample enters a microfluidic chip whose microscopic channels carry the liquid. One approach is to pinch the aqueous stream into tiny droplets surrounded by an immiscible carrier fluid, so each droplet becomes an isolated compartment. The objective is to create compartments containing exactly one cell often enough for single-cell analysis to be practical.
Loading follows Poisson statistics, so you deliberately run dilute: many empty droplets is the price of a high fraction of single-cell droplets. The cell is never glued down — it stays suspended in its own microscopic liquid environment.
Imaging
The compartments move through an imaging region where a microscope and camera capture repeatedly. Instead of one photograph per cell, Cyanoflow is built around time-series observation — the same compartment seen at t0…tn. That is what lets you measure behaviour, not just appearance.
Computer vision
Raw microscope frames are not useful to a model until they become measurable information. Segmentation and tracking convert pixels into per-cell numbers across four families: morphology, growth, optics and behaviour. The unit of storage stops being a file name and becomes a record.
- area
- perimeter
- shape
- aspect ratio
- structural features
- size change
- division events
- growth trajectory
- brightness
- pigmentation
- intensity distribution
- movement
- morphology change
- response over time
AI phenotyping
The model takes those measurements and searches for structure — clusters, outliers, and combinations of features that track with a phenotype you care about. Ten thousand cells collapse into a ranked shortlist you can actually work through by hand.
The model never gets to declare "this cell is good." You define what interesting means as a measurable rule — e.g. a combination of growth behaviour and optical phenotype — and the model finds cells that satisfy it.
Candidate recovery
Once a promising compartment is identified you want it physically separated from the rest for downstream characterisation — regrowth, sequencing, biochemical assay. This is where Cyanoflow stops being an analysis instrument and becomes a discovery platform.
Routing a specific compartment — dielectrophoretic or valve-based deflection triggered by the classifier — is the hardest subsystem. A first prototype can stop at reliable imaging, tracking and classification and still be a substantial project.
What makes it different
The difference is not the microscope and it is not the model. It is the resolution of the biological search: you are converting biological diversity into searchable data, then querying it.
- culture
- measure the average
- adjust conditions
- repeat
Population resolution. Rare biology is averaged away.
- individual cells
- measure each one
- identify unusual phenotypes
- investigate candidates
Single-cell resolution. Diversity becomes searchable.
Define what "interesting" means, then measure how well the rule performs
260 simulated cells with extracted features. Move the thresholds to set your candidate definition — precision, recall and accuracy are scored against held-out validation labels.
∧ pigment ≥ 0.60
| Cell | Area µm² | Growth | Pigment | AR | Condition | Classification |
|---|
How many predicted candidates really are candidates?
How many of the actual candidates did the system find?
How often is the classification correct overall?
Does re-imaging the same cell give the same measurement?
Every cell becomes a record — and the records become the asset
The target is to collect structured measurements from many individual cells, then query relationships across a population. Neither throughput nor measurement quality has been demonstrated for Cyanoflow.
This is the point where the platform stops being "a microscope with an AI attached." The queryable population — not any single image — is what lets you ask which cellular characteristics actually travel with the phenotype you are chasing.
One instrument, seven subsystems, two outputs
Normal cells are logged and pooled. Candidate cells are routed out, characterised, and folded back into the dataset as labelled biology.
Imaging, tracking and phenotype classification, validated for reproducibility. Automated recovery can come later.
Not "AI looked at algae" — but: can measurable single-cell features predict a phenotype with stated precision and recall?
Approved, well-characterised strains only; standard containment and disposal; no culturing of unknown environmental microbes.
A high-throughput biological discovery engine
Instead of manually examining hundreds of cells, the proposed system would examine individual cells at greater scale. Throughput and selection quality still need experimental validation.
Cyanoflow
DiscoveryProposed discovery workflow: single-cell compartmentalization, time-series imaging, feature extraction, candidate recovery and characterization. Not a validated capability.
Algaephyte
CultivationProposed cultivation workflow: sensors and growth models to study culture conditions. Cultivation performance and any handoff from Cyanoflow remain unvalidated.
Keeping the two technologies distinct is what makes the platform coherent: one machine that searches biology, another that grows it — instead of one machine trying to do everything simultaneously.
Research status.
The platform represents a developing research concept combining microfluidics, microscopy, computer vision, and machine learning for single-cell microalgal research.
Capabilities shown on this website — including every simulation and interactive console on this page — should be presented as design goals, prototypes, or research objectives unless experimentally validated.
For environmental or unknown microorganisms, experimentation will follow applicable biosafety, institutional, and science-fair requirements.
How we think, in writing.
Single-cell methods and shared imaging notes. Cultivation models and bounded control belong to the separate Algaephyte concept.
Discover what the population average hides.
Cyanoflow is being developed to make individual microalgal cells measurable, searchable, and experimentally testable.
Explore Cyanoflow ResearchUnder development · Research objectives · Not validated claims