Accelerating Strain Screening through Transfer Learning
DataHow and Procelys by Lesaffre worked together and managed to reduce experimental workload by up to 50% while improving prediction accuracy across scales and conditions.
Read the full case study to learn what made this possible.
Challenge
Can we learn scale dependency from historical data and predict how a new strain or yeast extract will behave at large scale based on small-scale data?
Procelys by Lesaffre screens many yeast extracts on hundreds of strains under various conditions and scales, each year, in order to provide the best yeast nutrient to fit needs of different microorganism. To begin this study, we focused on a dataset containing four strains evaluated with two NuCel® yeast (YE1 and YE2) extracts at two scales.
Objectives
OBJ. 1: Evaluate knowledge transfer potential: Using data from both scales, two NuCel® yeast extracts, and three strains, determine what data is required for the fourth strain to predict its behavior with both yeast extracts at large scale.
OBJ. 2: Increase process capacity: Assess whether the model can replace some experiments, reducing the number of runs per strain and yeast extract at both scales.
Should you wish to further explore the case with the DataHow team, please contact us directly.
