IUPred

What it is

Proteins are chains of small building blocks called amino acids. Many chains fold into a settled three-dimensional shape, but some stretches never do: they stay loose and keep changing form. These are called disordered regions. IUPred estimates where they are using only the order of the amino acids in the chain, with no experiment and no picture of the protein. For each position in the chain it works out how favourably that amino acid could bond with the others around it, giving an estimated interaction energy. Positions whose neighbours offer little to bond with score as disordered, because there is nothing to hold a fixed shape together.

The method comes in two settings, usually called short and long. The short setting is tuned to brief floppy patches, such as the loose ends of a chain. The long setting looks for extended disordered stretches running through a protein. Running both gives a fuller picture than either alone. The output is a score for every position, which can be summarised as the share of a protein predicted to be disordered, or drawn as a profile along the chain. Because it needs only the sequence, it can be pointed at a single protein or at thousands at once.

Why scientists use it

Disorder matters. Floppy regions let proteins change partners, switch on and off, and gather together into droplet-like clumps inside the cell. Faults in such regions are linked to disease. But the usual ways of working out protein shape, such as crystallising a protein and shining X-rays through it, tend to fail exactly where the protein will not hold still, so these regions are often missing from the structures on record. A sequence-based predictor fills that gap cheaply.

IUPred therefore stands in for an experiment that is hard to do. It is also one voice among several. Researchers commonly run a set of different disorder predictors over the same sequences and treat the regions where most of them agree as the trustworthy ones, since each method was built on different assumptions. IUPred's appeal in that company is that its reasoning is physical, resting on bonding energy rather than on patterns learned from past examples.

In the record

Limits

It is a prediction, not an observation. The score says a region is unlikely to hold a fixed shape, not what that region actually does, nor what shape it may take when it meets a partner protein. Some stretches fold only on binding, and those can be called disordered. The short and long settings can disagree, and so can rival predictors, which is why agreement between several is usually required. Results also depend on which version of a protein's sequence was used.

Written by AIxSci: the background from general knowledge, the examples from checked records.

Every article