Smarter Clone Selection: Earlier, more informed decisions for novel biologics 

For a drug developer, the clone is one of the longest-lived decisions in CMC. It becomes the cell substrate, sets the production ceiling, shapes clone-dependent product quality attributes, and sits on the critical path to the clinic. A better-informed clone decision, made sooner, protects both the program timeline and the quality of every downstream dataset. 

How clones are selected today 

The workflow is broadly the same across CDMOs, specialist CLD providers, and biopharma of every size: generate a stable pool, clone single cells with imaging to document clonality, expand through well plates on growth- and titer-based cuts, screen a short list in small-scale fed-batch, and pick the lead on growth, titer, product quality, and stability. 

Advances in molecular engineering and single-cell isolation have significantly accelerated cell line development, though the criteria used to select robust, stable, high-producing clones have changed the least. Productivity remains the main gate, and early static, unfed cultures are a poor stand-in for fed-batch performance. Machine learning applied to early single-cell cloning data is emerging, but most workflows today use early images after single-cell deposition to document clonality. 

At Wheeler Bio, we asked whether those existing images could tell us more, and the answer was yes. Clones that became the best producers were not necessarily the ones that a conventional screen would have favored. 

From clonality images to production forecasts 

Partnering with TEK Optima Research, we utilized their DeepInsight™ AI platform to turn well image series into quantitative growth traits. We mapped these traits on clones expressing a single IgG1 monoclonal antibody molecule in one CHO lineage, sampled the clone universe to span the full range of behavior, performed fed-batch bioreactor production, and then asked if data derived from early outgrowth of clonal isolates had predictive power for the outcomes of fed-batch production.  

We utilized a multivariate approach to understand how the clone outgrowth characteristics translated to bioreactor performance. The image-derived traits were combined with titer measurements during outgrowth into a small number of composite scores. Each score was then tested for its ability to predict fed-batch outcomes in clones held out of the model, not simply whether there was a correlation. 

Beyond predicting individual outcomes, we asked whether clones naturally fell into groups and whether these groups behaved alike during bioreactor production. Three distinct production phenotypes emerged, and the grouping proved statistically robust. Where a clone fell on this map of early outgrowth behavior predicted the kind of producer it became. 

The results permit clone down-selection grounded in the biology that drives production, rather than in static growth and titer snapshots alone. It can also happen earlier, using images already collected for clonality. Down-selection removes fewer strong clones, and time and resources focus on clones with the strongest predicted production performance. No new wet-lab assays, consumables, or instruments are required. 

For a drug developer, this turns clone selection from a leaderboard into a choice. A titer ranking tells you which clone produced the most in one experiment. A phenotype tells you how a clone is likely to behave: how it grows, what it consumes, and how it holds up over a production run. Because a clone’s position predicts the kind of producer it becomes, a program can match clone selection to its priorities and add another dimension to risk management. 

See the data 

We presented the full results at BioProcess International in September 2026, including which early traits matter, what they predict, what they don’t, and how far upstream the selection decision can move.