Hybrid fold recognition: combining sequence derived properties with evolutionary information

Recent assessments of structure prediction have demonstrated that (i) although fold recognition methods can often identify remote similarities when standard sequence search methods fail, the score of the top-ranking fold is not always significant enough to allow a confident prediction; (ii) the use of structural information such as secondary structure increases recognition accuracy; (iii) modern sequence-based methods incorporating evolutionary information from neighboring sequences can often identify very remote similarities; (iv) there is no one single method that is superior to other methods when evaluated over a wide range of targets, and (v) extensive human-expert intervention is usually required for the most difficult prediction targets. Here, I describe a new, hybrid fold recognition method that incorporates structural and evolutionary information into a single fully automated method. This work is a first attempt towards the automation of some of the processes that are often applied by human predictors. The method is tested with two cases that are often applied by human predictors. The method is tested with two fold-recognition benchmarks demonstrating a superior performance. The higher sensitivity and selectivity enable the applicability of this method at genomic scales.