Optimization analyses of Velvet algorithm based on RBF Neural Network

Abstract

Velvet is a very effective de novo assembly algorithm specifically designed for assembling read data from next generation sequencing platforms. Velvet runtime parameter “Hash Length” essentially affects the performance of assembly. This study proposed an effective method to resolve the problem that determination of optimal hash length greatly depends on the experience of the user. Firstly, we analyzed the effect factors of optimal hash length, including depth of coverage, base calling error rate and complexity of read data. Then, we set up a RBF (Radial Basis Function) Neural Network trained by various assembly data sets, which could automatically suggest optimal hash length for Velvet algorithm based on the effect factors. The experimental results proved the validity of our method.