A simulation toolkit for upstream bioprocess optimization and troubleshooting

Powered by hybrid ML-mechanistic models that predict the evolving metabolic state of cells.

Get in Touch See how we've helped
Experiment-based bioprocess design has limitations:
  • 01 Low Experimental Throughput
  • 02 Observations Without Causes
  • 03 Undetected Sensor Errors
  • 04 Biological Complexity

Traditional DoE can only test a small fraction of possible designs due to time and cost constraints.

Process data reveals little about intracellular mechanisms.

Sensor measurements may be unknowingly erroneous.

Predictive accuracy is hindered by the complexity of the biological system.

Vast design space
  • Experimentally tested
  • Possible designs
Anomaly without mechanism
  • Unexplained by-product rise
  • Process observations
Undetected sensor errors
  • Undetected error
  • Observed sensor readings
  • True values
Biological complexity
  • Obfuscated and complex interactions
Our digital twin addresses these limitations

A digital twin can explore thousands of configurations at a fraction of the experimental cost and time.

Digital twin bioreactor
  • Extracellular Module
    Reactions outside the cell that occur independent of cellular activity.
  • Metabolic Module
    Hybrid mechanistic-machine learning model that simulates reaction kinetics and cell growth.
  • Reactor Module
    Influxes, outfluxes, and reactor characteristics.
  • Identifysensor anomalies.
  • Troubleshootwith metabolic flux analysis.
  • Optimizetowards multiple competing goals.
How we model metabolism
01

Assemble a Genome-Scale Metabolic Model (GEM).

Genome annotation
Gene–protein–reaction connection
Cell with metabolic networks
Gap filling and curation
Experimental validation
02

Decompose into minimal metabolic functioning units.

Decomposing metabolic network
03

Train a neural network on process data to minimize prediction error by modulating the activity of those metabolic units.

Model training

12 of the top 20 pharma have used our digital twins for process optimization.

Reduced peak NH4+ concentration by 50% whilst keeping titer within 98.7%.
Experimentally validated
Improved product titer by 70%.
Experimentally validated
Improved key glycoforms by up to 91% in a partner's mAb.
Experimentally validated
Lifted titer constraints by supplementing key vitamins.
Simulation result
The Insilico Suite bundles model training, predictions, and results analysis into a no-code workstation.
Book a tour