This paper describes some useful applications of Design of Experiments methods to biofactory R&D. However, there are also some very serious issues in the paper. It is of particular concern because the issues are related to the claims made by the authors.
Starting with minor issues:
- Figure 1. Shows ‘Statistical machine learning’ but there is no clear description of what machine learning methods they used in the content of the paper.
- The caption for figure 2. appears to be misplaced to the bottom of page 5.
Moving onto major issues. The authors claim they reduced a design space of 4x3x3x3x24=2592 down to 16 representative constructs. This is a highly suspicious claim. The reasons are as follows.
- The authors claim there are four levels of expression by vector backbone (4 different backbones). However, in Figure 2. There is no vector backbone selection shown at all.
- In Figure 2.a ColE1 origin is never used, which suggests it is not actually used in the initial design space.
- There is a lack of orthogonality between representative designs as described below.
For the reasons above it’s difficult to confirm the author’s claims. The data presented in the paper conflicts their compression claim and description of methods they used.
‘Origin’ and ‘R0 promotor’ columns satisfy orthogonality by the number of combination repetitions:
p15A, Ptrc = 4
p15A, PlacUV5 = 4
pSC101, Ptrc = 4
pSC101, PlacUV5 = 4
However, R2 and R3 Promoter columns don’t meet this requirement, which can be observed at a glance.
