Iterative sequential Monte Carlo algorithm for motif discovery

The discovery of motifs in transcription factor binding sites is important in the transcription process, and is crucial for understanding the gene regulatory relationship and evolution history. Identifying weak motifs and reducing the effect of local optima, error propagation and computational complexity are still important, but challenging tasks for motif discovery. This study proposes an iterative sequential Monte Carlo (ISMC) motif discovery algorithm based on the position weight matrix and the Gibbs sampling model to locate conserved motifs in a given set of nucleotide sequences. Three sub-algorithms have been proposed. Algorithm 1 (see Fig. 1) deals with the case of one motif instance of fixed length in each nucleotide sequence. Furthermore, the proposed ISMC algorithm is extended to deal with more complex situations including unique motif of unknown length in Algorithm 2, unique motif with unknown abundance in Algorithm 3 (see Fig. 2) and multiple motifs. Experimental results over both synthetic and real datasets show that the proposed ISMC algorithm outperforms five other widely used motif discovery algorithms in terms of nucleotide and site-level sensitivity, nucleotide and site-level positive prediction value, nucleotide-level performance coefficient, nucleotide-level correlation coefficient and site-level average site performance.