fig3
From: Integrating sequence and chemical insights: a co-modeling AI prediction framework for peptides
Figure 3. (A) Average attribution values of each amino acid in the AP prediction task. The x-axis represents the types of amino acids and y-axis shows the average attribution scores from the sequence-based (in yellow) and graph-based (in green) models; (B and C) Variation in attribution results for models with different random seed initializations for the Transformer (B) and the GraphSAGE (C) architectures. The results indicate that different initializations of the learnable parameters have minimal impact on the consistency of the attribution results. AP: Aggregation propensity; GraphSAGE: graph sampling and aggregation networks.
