Decision story
Select the right CHO clones before the next expensive run
The expensive mistake is advancing clones that look good by titer but carry hidden waste burden, fragile growth, or weak model consistency. This demo shows the decision change, then exposes the evidence behind it.
Page 1 of 8
The screen
What data do we have for each clone?
The starting screen
50 CHO clones over 7 measured days
We begin with the kind of data a team already has after a clone screen: viable cell density, product, feeds, and metabolite measurements across replicate cultures.
Metric direction explorer
Rates
Clone
Interval calculations
Each interval uses the next measured timepoint
Metric directions
Compare clones before any ranking decision
Raw measurements over time
All 50 clones are visible; the selected clone is highlighted
Start by looking for trajectory shape, not just endpoint product. Strong endpoint titer becomes less attractive when it arrives with late viability loss, high waste, or heavy nutrient demand.
High titer is favorable only if it is not coupled to high overflow burden.
qP separates clones that truly produce more per viable cell from clones that mainly grew more biomass.
Screen-only view
What would we pick before the CHO model check?
Shortlist
Clone ranking after the CHO model check
Selection comparison
Screen-only, CHO model, and model-reviewed top 5
Detailed feature table
All derived values used by the learning page
Starting score
Edit Score v1
Target prediction
Move the weights and watch prediction quality
Formula search
Which formulas explain the selected target?
Score update
Best formulas become subscores
Detailed learning log
What each cycle tested
Shortlist update
Initial score vs updated score
Model inputs
Measured rates become per-clone model bounds
CHO model check
Model consistency across all clones
Next run
Five clones, one monitoring plan
Evidence package
What supports the decision?
Next step
Discuss a clone-selection review with real data
The strongest use case is a team with one expensive next run, enough screen data to calculate rates, and a need to justify a smaller, better-monitored shortlist.
Formula drawer and units
sum((VCD_start + VCD_end) / 2 * days_between_samples)Area under the living-cell curve. This tells us how many viable cells were available to make product.product made / IVCDProduct made per living cell per day. This separates true productivity from simply having more cells.start glucose + feed added - end glucoseIn fed-batch data, feed additions must be counted before calling a concentration drop consumption.concentration change / interval IVCDUptake and secretion rates are calculated between measured days, then summarized for the production window.pmol/cell/day -> mmol/gDW/hUses the explicit dry-cell-weight assumption so process rates can be displayed as model bounds.measured rate +/- 20%Bounds use ranges because exact constraints would overstate measurement precision.Assumptions and limitations
Assumptions: dry cell weight 330 pg/cell, mAb molecular weight 150 kDa, uncertainty bounds +/-20%, feed represented as mM-equivalent added since previous sample.
Limitations: this v1 does not run a full CHO genome-scale model in the browser. It does not claim exact intracellular flux prediction from endpoint data. The score improves with amino-acid panels, oxygen and CO2 signals, replicate metadata, and real process context.
Current FBA layer: the demo uses CHO.net 1.2 with native CHO exchange reactions and a native mAb product reaction to review measured clone behavior. For a production scientific package, replace or extend the compact model with the approved CHO model for the specific cell line and process.
Raw long table