Fig. 4: Comparative analysis of misclassification rates in the pathogenicity detection task.
From: Optimized model architectures for deep learning on genomic data
The baseline models are shown in gray, while the red bars indicate the models developed by GenomeNet-Architect. The data for the dataset itself and the baseline results, with the exception of DNABERT33, were derived from the DeePaC study13. In addition, the pre-trained DNABERT33 model is fine-tuned on this task and added as a baseline. The graph shows individual model performance along with the improved performance archived by the ensemble approaches and highlights the superior performance of the GenomeNet-Architect models over various baselines.
