Skip to main content
. 2024 Dec 2;25(23):12942. doi: 10.3390/ijms252312942

Figure 2.

Figure 2

Accuracy of four different combination strategies for incorporating pre-trained embeddings, DNA2Vec and DNABERT, with the learnable 4-mer embedding. Panels (A–C) denote three different combination approaches for incorporating DNA2Vec embeddings into DeepEnhancer, where each bar represents a different value of k used for k-mer-based DNA2Vec embedding, ranging from 3 to 8. Panel (D) represents the performance of introducing DNABERT into DeepEnhancer, using the ‘direct-concat’ approach. The y-axis shows the accuracy of the validation dataset.