Question 4 of 6 Intermediate Scenario
How would you plan a design-build-test-learn cycle?
Short answer
Define the engineering objective, propose informative variants, build them reliably, measure them with controlled assays and use results to choose the next round. Each round should reduce uncertainty, not just generate more mutants.
Why
Start with a quantitative objective such as improved catalytic efficiency above a specified pH while maintaining minimum soluble yield. Use structural, sequence and prior assay information to propose a small diverse variant set with deliberate exploration and baseline controls. Confirm sequence identity after construction, measure expression and soluble recovery, then assay function in the intended operating range with replicates.
Store variants, conditions, raw data and failed builds, not just top-ranked hits. Fit an interpretable or probabilistic sequence-function model if data volume supports it, and use the uncertainties to balance exploitation against informative experiments. For example, two mutations may look promising alone but display antagonistic epistasis together, a result that should inform the next library.
Retest selected variants independently to avoid optimizing noise or assay artifacts. Stop or redefine the objective when gains are small, model predictions are unstable or a different property becomes the limiting constraint.
Common mistakes
- Choosing the next library solely from a model rank without verifying measured fitness and failure modes
The interviewer might ask next
How would you allocate a 96-variant budget between exploration, controls and known strong designs?