Detection of compositional constraints in nucleic acid sequences using neural networks

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Introns and exons are crucial components of eukaryotic genes, with distinct sequence characteristics.
• Previous computational methods often treated introns as negative examples for exon learning.
• Understanding compositional constraints within these regions is vital for gene structure and function analysis.

Purpose of the Study:

• To develop and evaluate a neural network method for detecting compositional constraints in introns and exons.
• To jointly learn intron and exon sequences using junk DNA as a common counter-example.
• To assess the method's performance in identifying and classifying gene elements.

Main Methods:

• A two-phase neural network algorithm: a learning phase using back-propagation and a generalization phase for testing.
• Joint learning of intron and exon sequences, with junk DNA serving as a negative reference.
• Application to human globin cluster sequences and classification of unknown examples.

Main Results:

• The neural network successfully discriminates between introns and exons, achieving correlation coefficients of 0.50 and 0.64, respectively.
• The use of junk DNA in learning enabled the detection of constrained regions within introns and exons.
• The method demonstrated effectiveness in classifying unknown DNA sequences.

Conclusions:

• The proposed neural network method offers a robust approach for identifying compositional constraints in introns and exons.
• Joint learning with junk DNA as a counter-example enhances the detection of sequence-specific organizational patterns.
• This technique holds potential for advancing the study of internal sequence organization in genes.