Evolving side effect machines to measure distances between DNA promoter sequences

Side Effect Machines are tools for extracting numeric features from alphabetic sequences. They have been used successfully on many DNA sequence classification tasks. This paper continues development of a new type of application for them, the creation of a distance measure between DNA sequences. The application problem is to find distances between promoter sequences of genes. These promoter sequences contain potentially dozens of short sequence motifs to which transcription factors bind. Finding relationships between promoter sequences has been addressed computationally in the past by looking for these sequence motifs. The method presented here has the potential to take into account not just individual motifs, but the entire genomic landscape. It leverages data on transcription factor binding sites determined with biological experiments on yeast and fruit flies to develop useful features.