We now discuss the feature engineering required to prepare bytes files for training and classification using Strand.
Next, we give a very short overview of the Strand classification process (see [15] for more details).
We tested the success rate of coding strand classification on the 5′ and 3′ sides of CDSs.
This highlights the fact that mutations can potentially affect expression of transcripts on both strands, and that the classification of a SNP is strand dependent.
Finally, we evaluated the performance of UFM by comparing it to the success rates of the classification of coding sequence, strand, and frame diagnosis of the largest ORF (LORF) of the sequence considered.
Processing this data in a fashion similar to the bytes files as raw text would result in many of the unique token values being broken across each gene sequence word created during Strand training and classification processing.
Clicking these tracks returns a page containing features of the respective RNA locus, such as anticodon and amino acid specification (for tRNA), unique family classification (for miRNAs and snoRNAs), strand information; complete sequence as well as 2-dimensional structure notation.
However, large performance gains are achieved in Strand by using binary classification techniques where no nested categorical frequency values or log based calculations are required during classification function operations.
