Unsupervised Pattern Discovery in Biosequences Using Aligned Pattern Clustering

Protein, RNA and DNA are made up of sequences of amino acids/nucleotides, which interact among themselves via binding. For example, (1) protein-DNA binding regulates gene transcription [1]; and (2) Protein-protein binding plays important roles in cell cycle control and signal transduction [2].The binding is maintained by either the direct participation or assistance of conserved short segments of biosequences called functional elements. Because of their importance in preserving function, they are well conserved throughout evolution. Their recognition is therefore essential for an in-depth understanding of the biological mechanisms [3] such as inhibitor design [4]. Although these functional elements could be discovered from the three-dimensional structural forms of the biosequences, the applicability is limited due to the high experimental cost. With the advent of new sequencing technologies [5], it is preferable to discover, directly from the abundant biosequence data, functional elements where many of them are short with variable length, like Short Linear Motifs (SLiMs [6] ) which play important roles in protein-protein interaction but are only 3 to 15 amino acids in length. Such short elements could not be captured well by the popular position weight matrices [7]. In this paper, we aim to briefly review an unsupervised pattern discovery tool known as Aligned Pattern Clustering (or its software WeMineTM) [8-11] which is developed to facilitate the discovery and analysis of patterns in biosequences. Its applications include 1) identifying functional elements in protein sequences [8-11,2] revealing functioning subgroup characteristics of functional elements [12-14,3] identifying co-occurring intra-protein [15,16], inter-protein [17] and proteinDNA functional elements [18,19].