Fig. 2

An analysis of existing deep-learning-based taxonomic classification methods and MetageNN for long-read settings. For each method, we a show the epoch run time elapsed and b display the inference time in reads-per-second. c, d and e show the results for F1 score on three test datasets simulating higher, moderate and lower rates, respectively. As compared to existing approaches, MetageNN has the shortest epoch run time, the highest speed and the highest F1 scores demonstrating its robustness to sequencing errors present in long-read data