Distinguishing between competing alignments of protein structures is still a difficult problem after more than 30 years. It comes down to optimising for two competing objectives:
- Maximise the number of residue-residue correspondences, and
- Maximise the similarity between the structures after superposition.
I-value, a new alignment evaluation method, leverages the inherent information content within protein structures by treating the alignment as a hypothesis on the similarity between structures. This allows the quality of an alignment to be quantified using the explanation message length.
Therefore, I-value has the following important properties:
- The difference in I-value between two competing alignments is their log-odds ratio,
- the Null Hypothesis is simply the raw encoding of the two structures separately (i.e. without the alignment/similarity hypothesis), and
- I-value is an objective trade-off between coverage and similarity without any arbitrary parameters of cut-offs.
I-rate refers to this server which can be used for two purposes. To calculate the I-value of a given alignment; or to decide which of two alignments is the best one. The mode of operation is chosen by selecting one of two radio buttons that appear at the top of the form.