DOE approaches can handle multiple variables
Full factorial design: for n variables requires 2n data points
- not practical beyond 6 variables
Use fractional factorial designs 2(n-x), where x is a trade-off factor
- high x gives low resolution but few data points needed
- interactions become confounded with main effects
- low x gives higher resolution but more data points (reagents) needed
Software can determine the best compromise
- Resolution IV usually adequate (resolves main effects from higher order interactions)
Can optimize for any desired outcome (s/b CV, stability etc)
Many software packages available (JMP, Design Expert, Statistica), but all give output as a spreadsheet -need to translate into robot instructions.