COMPUTATIONAL INFERENCE TECHNIQUE FOR MINING STRUCTURED MOTIFS

Abstract

One of the major challenges in bioinformatics is the development of efficient computational tools for mining patterns. Structured motifs, like DNA binding sites in organisms with peculiarities in their genomic sequence like malaria parasite, Plasmodium falciparum have not been mined by existing structured motifs extraction tools. There is a need to develop faster computational tools to mine these DNA binding sites which are viable drug targets. This work was aimed at developing an algorithm for mining structured motifs in the genome of P. falciparum. The Gene Enrichment Motif Searching (GEMS) method for mining simple motifs was modified by incorporating the time efficient implementation of the suffix tree data structure with suffix links. This enables an improved searching speed, while adding an optimized position-weight matrix computation using the hypergeometric-based scoring function. This algorithm, Suffix Tree Gene Enrichment Motif Searching (STGEMS) was implemented in C programming language on Linux platform. An empirical evaluation of the sensitivity of STGEMS was conducted by comparing the similarity check mechanism of the GEMS algorithm for mining simple motifs with that used in another popular algorithm for extracting structured motifs, a Multi-Objective Genetic Algorithm Motif Discovery (MOGAMOD). The output of STGEMS algorithm was validated by comparing the motifs discovered with those obtained using biological experiments. A further validation was done by applying the STGEMS and GEMS algorithm to selected metabolic pathways and the results were compared. The STGEMS algorithm was tested with four sets of genes from the intraerythrocytic development cycle of P. falciparum. The speed of execution was evaluated using three simple motif discovery tools: Expectation Maximization Motif Elicitation(MEME), Gene Enrichment Motif Search (GEMS), and WEEDER as well as two structured motif discovery tools: RISOTTO and EXMOTIF on four different gene sizes.The high sensitivity of STGEMS in mining structured motifs from sequences in P. falciparum was proven empirically by its ability to identify 91% of the motifs in the sequences while MOGAMOD could not identify any motif. This validated the high sensitivity of the similarity check mechanism employed, in contrast with that used in MOGAMOD. The STGEMS algorithm identified 90% of the binding sites in P. falciparum which were similar to those obtained in biological experiments. On the selected metabolic pathways, STGEMS discovered all the simple motifs identified by GEMS, in addition to the structured motifs which GEMS could not identify. The empirical runtimes of STGEMS, MEME, WEEDER, GEMS, RISOTTO and EXMOTIF were respectively 20, 35, 26, 25, 28, 30 seconds for 20,000 base pair (bp), 32, 43, 44, 45, 42, 40 seconds for 40,000 bp, 41, 55, 56, 55, 52, 50 seconds for 60,000 bp and 54, 68, 69, 65, 67, 61 seconds for 80,000 bp respectively. The proposition resulted in a linear asymptotic runtime of O(N) at each iteration of the algorithm. The suffix tree gene enrichment motif searching algorithm developed was time efficient and successful in mining structured motifs like DNA binding sites in Plasmodium 15 falciparum. This will aid a faster drug target discovery pipeline for the design of effective anti malaria drugs.