Fig. 1

Schematic of the steps to generate “bag-of-k-mers” and “vector-k-mers” models. The workflow shows the steps from data preprocessing to model output. We fitted “bag-of-k-mers” and “vector-k-mers” models for k values between 5 to 10 bp (within the common range in which regulatory elements have been observed). Training and evaluation of both methods happened on the same portion of the data to facilitate comparisons. The common pre-processing step involved the collapsing of complementary k-mers as the same token to reduce the noise of k-mer counts and the effective vocabulary for feature selection. The final outputs are both the classifiers and learned features