Design
Trade-offs: which candidates in a table are not beaten on every objective at once
Put a table of candidates in, say which columns should be high and which low, and see which ones no other candidate beats on every count at once. Everything else is placed on a deeper front with the candidate that beats it named, and the first two objectives are drawn against each other, coloured by front.
Which red monomers no other beats on brightness, maturation and pKa at once, from FPbase’s own measurements
What it does
- 01Which candidates in a table no other candidate beats on every objective at once: the first Pareto front
- 02Every other candidate placed on its front, with the candidate one front up that beats it named
- 03Each objective maximised or minimised, any number of them, in the table's own units
- 04Rows filtered first by any column (an emission range, a monomer, a state), and rows with a missing measurement left out and counted rather than ranked last
- 05The crowding distance along each front, NSGA-II's normalised form or raw gaps, so a short list can be spread along the front
- 06The first two objectives drawn against each other coloured by front (the first joined when there are only two), every candidate pickable
What it runs on
Pareto ranks and the non-dominated set are checked against moocore 0.3.2 on 230 cases and 49,957 points, every one identical, and against an independent fast non-dominated sort (pymoo 0.6.2). Crowding distances match emoa 0.5-3 and pymoo exactly on 1,651 fronts. The whole ranking is checked against R over 1,239 real fluorescent-protein states from FPbase. The panel runs on @ports/workflow.
Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.
Questions
- Which candidate on the first front should I pick?
- That depends on how much each objective matters to you, which the ranking does not assume. Every candidate on the first front is a defensible choice; the ones further down are each beaten by a named candidate on every objective.
- What happens to rows with a missing value?
- They are left out and counted, with examples. A missing measurement is not a bad one, so ranking it last would be wrong.
- Is anything uploaded?
- No. The ranking runs in the browser.
Also searched as
Pareto front · non-dominated sorting · multi-objective comparison · which candidate is best on several criteria · pick a fluorescent protein · brightest fastest maturing red fluorescent protein · NSGA-II crowding distance · Pareto optimal designs