RAPTOR's Approach to Prediction

Given a query protein sequence, RAPTOR scans a template libary which is a set of known structures derived from the PDB. For each structure template, RAPTOR threads(aligns) the query sequence against the template by optimizing a scoring function and an optimal alignment will be obtained. After threading, all the aligments are ranked by a statistical measure. The structure of the query sequence is built on the alignment from the top template.

The scoring function used in RAPTOR includes terms associated with:

• Sequence homology
• Secondary structure types
• Solvent accessibility
• Pairwise interaction

The weights of different terms in the scoring function are optimized by using a generic algorithm.

RAPTOR provides three different threading algorithms:

• No Core: Dynamic programming used to align a query sequence to a template (algorithm used in PROSPECT)
• Non-Paiwise Core: Dynamic programming used to align the query sequence to the template (template parsed as a series of cores connected by loops)
• Integer Programming (Patented by BSI): Integer programming used to align the query sequence to the template. Pairwise interactions are treated rigorously (most servers cannot do this)
• Pairwise interaction

Statistical measures used to rank aligments:

• An SVM (Support Vector Machine) technique is used to rank the alignments after threading. The resulting score reflects the quality of the sequence-structure alignment.
• RAPTOR employs a BLAST-like E-value to evaluate a sequence-template alignment, which provides an overall measurement of prediction quality.

RAPTOR Workflow

Here is an illustration of RAPTOR's work flow: