Figure 6
Results of the semi-supervised learning investigation using a deep learning model. The model was trained using data collected up to 2019. (A) The organizational structure of the chaos game. An example of the composition of the NAAA super-pixel can be found in the top left corner. (B) Saliency map for each FCGR. Highlighted are 4 × 4 super-pixels of k-mer frequencies corresponding to the different regions of the FCGR presented in (A). For example, the patch found in the top left corner represents a collection of k-mers ending in AAAA. High saliency regions are warmer and are used by the model to differentiate between sequences. (C) This table displays the results of the fivefold stratified cross-validation experiment. Predictions which were correctly made are found where both the column and row labels are identical. False negative predictions for each genus are found along the rows (eg: five Anopheles sequences were predicted to be Non-Culicidae dipterans) while false positives are found along the columns.
