01
Cell Localisation
Detects and locates every cell within brightfield well images. Precise cell-level positioning across the full imaging field.
RESEARCH
The Science of Prediction
DeepInsight uses a novel modeling approach to accurately map cell growth within confined environments, predicting later behaviours from early imaging data alone.
We are actively compiling data on label-free phenotypic clustering to link growth patterns with stability.
Read the published paper →Model Methodology
Four proprietary neural networks, designed and trained in-house. No third-party foundation models.
The team behind DeepInsight designed and deployed the VIPS single-cell printer and its production AI models — the system now used across hundreds of CLD laboratories worldwide. That same first-hand understanding of cell-line development imaging and machine learning at production scale now drives the DeepInsight pipeline. Every network is designed, trained, and deployed by engineers who have lived the CLD workflow.
02
Cell Categorisation and Measurement
Classifies cell types and measures morphological features. Builds a quantitative profile for each detected cell.
03
Counting, Summarisation and Growth Profiling
Aggregates per-well and per-colony cell counts. Tracks population growth over time and profiles growth dynamics.
04
Neural Fingerprinting and Clustering
Generates a unique optical fingerprint for every detected cell. Clusters clones by phenotype to surface the most stable, high-performing candidates.
Key models
Each model addresses a specific decision point in the CLD workflow. Together they build a complete clone-level profile, helping you select the best candidates with confidence.
Growth prediction
Exponential sigmoid growth model splits each clone's trajectory into early passive and later dominant growth components. From the first few days of culture, the model predicts which clones will sustain robust growth through to the end of the workflow.
Read the paperLabel-free cell counting
Cell counts without stains or fluorescent labels. Validated against manual counting and fluorescence methods at production scale. Per-well and per-colony counts feed directly into growth rate calculations.
Clone ranking and fingerprinting
Multi-factor ranking across growth rate, recovery speed, and colony density. The ranking surface flags the highest-potential clones for progression while catching underperformers before they reach the banking stage.
Validation
Trained and validated against production CLD data at scale.
Manual cell counting by experienced CLD scientists
Fluorescence-based viability assays
End-point titre measurements for growth rate validation
Cross-referenced growth profiling and additional instrument endpoints
DeepInsight models have been trained and validated across thousands of production plates — representing hundreds of thousands of wells — from collaboration partners and internal datasets. Validation metrics span cell count accuracy against manual counts by experienced CLD scientists, cross-referenced against fluorescence-based viability assays and end-point titre measurements. Growth profiling has been independently corroborated against additional instrument endpoints. The result is a model suite that performs reliably across diverse instrument stacks, cell lines, and laboratory conditions.
Data and intellectual property
Your data stays yours.
Customer data isolation
Your imaging data is never used to train models for other customers. Only data shared with us through partnerships is used for model training. Your data stays yours.
UK infrastructure
Purpose-built AI compute facility in Dorset, UK. GPU-powered inference on our own processing and storage hardware. GDPR-compliant. No external cloud providers.
Geographic expansion
Additional geographic regions planned, with USA processing capability initially.
Notify me when the research is published.
See how these models deliver decision-grade analytics in production.
Explore the platform