Optimization profiles

Windowed GC content and codon adaptation before and after an optimisation, on the sequence's own coordinates — so an improvement in the mean can be checked for where it actually happened.

What it doesCopy Anchor Link

An optimisation reports two numbers: GC before and after, CAI before and after. Both are averages over the whole region, and an average can improve in ways that do not help. A run that lifts codon adaptation across most of a gene while leaving one 60-codon stretch unchanged reports a better CAI and still contains the stretch that will slow translation. A run that pulls the mean GC into range by overshooting at one end and undershooting at the other reports a compliant number over a sequence with a GC-rich block in it.

These are the same quantities computed in a sliding window and drawn against position, with the original and the optimised sequence as two lines on one axis. Where the lines separate is where the optimiser worked. Where they coincide is where it did not — either because nothing needed changing, or because a constraint prevented it.

Both profiles are plotted on the document's own base-pair coordinates rather than on an index into the region, so they line up with the map's ruler and with each other. A coding sequence on the bottom strand is mapped through the same transform, and reads left to right under the same ruler as everything else rather than backwards.

When you'd use thisCopy Anchor Link

• After any run, before you order the sequence. The summary tells you the run succeeded; the profile tells you what it produced.
• When a constraint is guaranteed rather than measured. A GC bound the solver enforced by only ever emitting compliant bases is satisfied everywhere by construction — the profile is where you see how much room it had.
• When expression is still poor after optimisation. A residual trough in windowed codon adaptation, at the 5′ end especially, is a candidate explanation that a mean cannot produce.
• To choose a window that matters. A local dip that survives at every window size is structural; one that vanishes as the window widens was a single rare codon.

What this computesCopy Anchor Link

• Windowed GC content, the fraction of G+C in each sliding window, for the original and the optimised sequence over identical windows — so the two curves are directly comparable point for point rather than being two independently-chosen samplings.
• Windowed codon adaptation index, the geometric mean of the relative adaptiveness weights (Sharp & Li, 1987) of the codons in each window, for the host codon usage table you selected. The window is in codons and is mapped to base pairs for plotting, so it sits on the same axis as the GC profile.
• The window and step actually used, chosen from the length of the region — a short region gets a short window, because a window comparable to the region reports one point and no shape.
• The mapping onto document coordinates, including strand, so every position on these plots is a position you can go to on the sequence.

What it does not doCopy Anchor Link

A windowed index is a description of the sequence, not a prediction of expression: codon adaptation correlates with expression across genes and does not determine it for any one gene. Neither profile models mRNA folding, ribosome pausing, or the effect of the untranslated regions — the optimiser's own constraints handle structure separately, and these two curves say nothing about it.