To find a more accurate and efficient way to separate human and mouse reads, we next utilized Convolutional Neural Networks (CNN),42 which was originally developed for computer visualization and image processing.52, 53 CNN performs effectively in many natural language processing (NLP) problems including sentence modeling, text classification and query processing.54–58 We applied CNN here to classify RNA sequencing reads of different species using the identified features directly from the fragments without pre-aligning to reference genome.
