HYBRID MODELING · 2024
Predicting cell growth, titer and glucose demand
A CHO study combines process knowledge and data-driven modeling to predict culture performance across selected conditions.
THE SCIENTIFIC QUESTION
Can a hybrid model predict culture performance as process conditions change?
How the study approaches it
The authors combine mechanistic and data-driven components and compare modeling approaches for mammalian antibody production.
REPORTED FINDING
The authors report predictions of viable cell density, titer and cumulative glucose consumption under varying conditions, including a different product under similar process conditions.
What the evidence supports
Reported performance applies to the studied systems. It does not establish reliability for every cell line, scale or quality attribute.
CONNECTION TO BIOMEDANA
BioMedAna's CHO model also combines mechanistic kinetics with a data-driven correction, retained for selected outputs when it improves held-out error. This is a related approach, not a reproduction of this paper.
Explore BioMedAna Twin ↗ANALYSIS PROTOCOL
Decision, data and execution method
Compare a temperature choice under the same nutrient-addition policy. Read growth, substrate and product trajectories together rather than selecting an operating point from final titer alone.
Input data and units
| Input | Value / range | Role in the calculation |
|---|---|---|
| Time | 0–14 days | Daily reported samples; integration between samples |
| Viable cell density | 10⁶ cells/mL | Initial 0.5; modeled capacity 30 |
| Glucose | g/L | Initial 5; addition is g/L/day |
| Product | g/L | Initial zero; accumulated specific productivity |
| Temperature | 33–37 °C | Selected scenario; reference is 37 °C |
Analysis sequence
- Start both scenarios with identical states and glucose addition.
- Calculate temperature-dependent growth and productivity plus glucose-limited growth and death.
- Integrate three coupled balances using RK4 at 0.02-day steps.
- Report daily states, final titer, peak cells and total glucose addition.
- Inspect glucose depletion and cell loss before interpreting a titer difference.
Calibration and evaluation
Coefficients are assumptions, not a regression fit. There is no training dataset, measured validation error or fitted hybrid correction. Solver convergence is tested by halving the step. Deployment would require fitting on completed runs and scoring each output on independent held-out runs.
Reading the results
- Cooling can increase assumed specific productivity while slowing growth. Coupled trajectories reveal the net effect.
- Addition is amount per constant volume, not pump flow. Scale-up requires feed-concentration and volume balances.
- Product concentration is not recovered yield or a product-quality prediction.
Next experiments and deployment evidence
- Collect aligned cell, glucose, titer, temperature and feed measurements.
- Separate cell-line changes from operating conditions and measurement variability.
- Evaluate residuals by time and condition; verify novel scenarios experimentally.
ASK ASSIST · GUIDED EXAMPLE
Explore the question in BioMedAna's scientific workflow.
Choose a request to execute the educational model and inspect the results. This demonstration uses predefined requests and local calculations; it is not connected to the production Assist service.
Compare 34.5 °C with the 37 °C baseline
Compare 34.5 °C with the 37 °C baseline
BIOMEDANA WORKED EXAMPLE
CHO culture: growth, glucose and product trajectories
A reduced mechanistic example derived from the structure of BioMedAna's CHO engine. Compare a 37 °C baseline with a selected temperature under the same glucose-addition rate.
Viable cell density 10⁶ cells/mL
Product titer g/L
Glucose concentration g/L
Calculated live in your browser from the equations below. These are BioMedAna illustrative outputs, not experimental observations or published study results.
Model structure
μ = 0.8 · exp[−0.10(37 − T)] · G/(0.2 + G) · (1 − X/30) kd = 0.03 + 0.4 · 0.2/(0.2 + G) dX/dt = (μ − kd)X dG/dt = u − (μ/2 + 0.01)X dP/dt = 20 · [1 + 0.05(37 − T)] · X · 10⁻³
Parameters and initial conditions
All values below are assumptions for this worked example, not parameters attributed to the paper.
| Parameter | Value | Unit / basis |
|---|---|---|
| Initial viable cells | 0.5 | 10⁶ cells/mL |
| Initial glucose | 5 | g/L |
| Initial product | 0 | g/L |
| Cell capacity | 30 | 10⁶ cells/mL |
| Maximum growth rate | 0.8 | day⁻¹ |
| Glucose half-saturation | 0.2 | g/L |
| Specific productivity at 37 °C | 20 | pg/cell/day |
Interpret the example
Temperature changes both growth and specific productivity in these assumed equations. A higher specific productivity does not necessarily produce a higher final titer when fewer viable cells accumulate.
Assumptions and limits
Illustrative parameters; constant volume; no ammonia, lactate, dead-cell, oxygen or quality states. No fitted data-driven correction is included here. This is not the paper's hybrid model or a qualified BioMedAna prediction.
The culture examples use fourth-order Runge–Kutta integration with 0.02-day steps and non-negative state bounds. They simulate glucose addition at constant volume. No uncertainty band is shown because this example has no validated error model.
Inspect the implementation
const mu = 0.8 * Math.exp(-0.10 * (37 - temperature))
* glucose / (0.2 + glucose) * (1 - cells / 30);
const kd = 0.03 + 0.4 * 0.2 / (0.2 + glucose);
const qp = 20 * (1 + 0.05 * (37 - temperature));
const dCells = (mu - kd) * cells;
const dGlucose = feed - (mu / 2 + 0.01) * cells;
const dProduct = qp * cells * 0.001;
// Integrate with RK4, dt = 0.02 day.Run locally by importing runStudy from the downloaded module. The displayed snippets explain the implementation; the download contains the complete engine.
Inspect calculated values
| Day | Cells (10⁶/mL) | Glucose (g/L) | Product (g/L) | Added glucose (g/L/day) |
|---|---|---|---|---|
| 0 | 0.500 | 5.000 | 0.000 | 0.0 |
| 1 | 0.864 | 7.297 | 0.015 | 2.5 |
| 2 | 1.491 | 9.449 | 0.041 | 2.5 |
| 3 | 2.543 | 11.367 | 0.085 | 2.5 |
| 4 | 4.228 | 12.930 | 0.160 | 2.5 |
| 5 | 6.748 | 14.020 | 0.282 | 2.5 |
| 6 | 10.141 | 14.591 | 0.470 | 2.5 |
| 7 | 14.124 | 14.764 | 0.743 | 2.5 |
| 8 | 18.110 | 14.825 | 1.106 | 2.5 |
| 9 | 21.512 | 15.074 | 1.553 | 2.5 |
| 10 | 24.046 | 15.677 | 2.067 | 2.5 |
| 11 | 25.750 | 16.639 | 2.629 | 2.5 |
| 12 | 26.822 | 17.885 | 3.221 | 2.5 |
| 13 | 27.468 | 19.325 | 3.833 | 2.5 |
| 14 | 27.851 | 20.887 | 4.455 | 2.5 |
Before applying it to your process
Define the intended decision and acceptance criteria. Evaluate the method on representative held-out runs, inspect assumptions and uncertainty, and confirm performance prospectively before relying on its recommendations.
Editorial review: October 13, 2026. This report combines a publication summary with a BioMedAna educational implementation. The simulated results belong to the worked example; they do not reproduce the published experiments or establish a product benchmark.
EXPLORE MORE RESEARCH