Hard problems · Biotech
Choosing the next experiments instead of guessing them
A working session on a biotech company's experimental data, to identify the most promising parameters to test next.
The problem
In biotech R&D, developing a product or a protocol means searching for the right combination of parameters. Each combination is tested through an experiment, and each series of experiments costs time and consumables. Teams rely heavily on experience and intuition, while the results of past series remain largely untapped.
The approach
Structure existing experimental results: parameters tested, conditions, measurements.
Model how the parameters affect the target performance, and check that the model correctly predicts experiments it has not seen.
Estimate the uncertainty of each prediction, to tell solid leads from risky bets.
Propose the next parameters to test, each with its estimated chance of beating the best result so far.
The result
An understanding of what drives performance, and a targeted next series of experiments. Experimental validation is underway.
What's transferable
Cosmetics formulation, industrial process tuning, food product development: wherever progress comes through costly experiments, past trials hold information. A rigorous method helps choose the next trials, and so reach the right setting in fewer rounds.