SOMEA: self-organizing map based extraction algorithm for DNA motif identification with heterogeneous model.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate discrimination of transcription factor binding sites (TFBS) from background DNA sequences is crucial for computational motif discovery.
• Existing clustering algorithms often use a homogeneous model, assuming similar characteristics for both motifs and background signals, which limits their effectiveness due to distinct signal properties.

Purpose of the Study:

• To develop a Self-Organizing Map (SOM) based clustering algorithm for enhanced extraction of binding sites in DNA sequences.
• To improve the discrimination of motifs from background signals by employing a novel intra-node soft competitive procedure with adaptive weighting.

Main Methods:

• Developed a SOM-based clustering algorithm incorporating an intra-node soft competitive procedure.
• Utilized an adaptive weighting technique applied to two distinct signal models to differentiate motifs and background sequences.
• Compared the proposed method against existing motif discovery tools, including SOMBRERO, using real and artificial datasets.

Main Results:

• The proposed algorithm achieved significant improvements in average precision rates (approximately 27%) on real datasets compared to SOMBRERO, a state-of-the-art SOM-based tool.
• The method maintained sensitivity while enhancing precision, demonstrating superior performance against other motif discovery tools.
• Validated the effectiveness of using heterogeneous models for signal representation in motif discovery.

Conclusions:

• Motif discovery frameworks utilizing model-based clustering should adopt heterogeneous models to accurately represent distinct signal classes (motifs and background) in DNA sequences.
• Heterogeneous models offer superior signal discrimination capabilities compared to traditional homogeneous models, leading to more accurate motif identification.