Skip to main content
← All resources

Case study · With Procelys by Lesaffre

Accelerating strain screening through transfer learning

Can scale dependency be learned from historical data, so a new strain’s large-scale behavior is predicted from small-scale data? A study with Procelys by Lesaffre.

  • Microbial
  • Strain screening
EXPERIMENTS PER STRAIN VS STANDARDBIOLECTOR XT−38%AMBR250−50%STRAINS PREDICTED AT LARGE SCALE4 PROBIOTIC STRAINSEACH PREDICTED ONCE
  • 38%

    Fewer experiments at BioLector scale

  • 50%

    Fewer experiments at ambr250 scale

  • Improved prediction accuracy across scales and conditions, without compromising decision quality

The challenge

Small and large scale often disagree

Strains and yeast extracts are screened at microliter scale to find the best-performing combination, then a few are evaluated at 250 mL to 20 L. Performance at small and large scale often does not align.

The approach

Predict each strain at large scale

Using data from three strains across both scales and yeast extracts, plus small-scale data from the fourth, the model predicted the fourth strain’s biomass growth at large scale. This was repeated so every strain was predicted once.

  • 4 probiotic strains
  • 2 NuCel® yeast extracts
  • 2 scales

Get the full results

Download the full case study, with project outline, analysis, and case results.