Why a CDS-level pass does not guarantee expression
Codon choice, protein architecture, experimental context, and measured yield belong to connected—but different—evidence layers.
A clean result can answer the wrong question
A private case prompted a familiar but important question: if a coding sequence passes computational checks, why might the downstream experiment still be unsuccessful?
The answer begins by separating layers. A CDS review can test amino-acid identity, codon choice, nucleotide composition, configured motifs, and assembly-related constraints. Those checks can identify useful design risks. They do not directly measure protein folding, secretion, solubility, cellular burden, construct context, assay behavior, or final yield.
Three different statements
These statements should not be collapsed:
- The CDS satisfies its computational contract.
This is a reproducible software result. - The protein has no obvious architecture-level risk indicators.
This may be an annotation or advisory assessment. - The construct performs well in a defined experiment.
This requires an experimental observation with controls and provenance.
A pass at the first layer does not establish the third. The reverse also matters: an imperfect computational score does not prove that expression will fail.
The product boundary
The practical response was not to make the CDS score claim more. It was to preserve the CDS layer and add clearly labeled protein-risk annotations where useful. Such annotations remain warnings, not substitutes for experiments.
This is why public Eijex language distinguishes design review from biological validation. The software can produce reviewable candidates and traceable checks. Experimental systems determine what happens in the laboratory. Linking the two is valuable, but linkage does not erase the boundary between them.