Fig 3.

(a) shows the experimental results of ablation of two groups of feature encodings on SMFM, where the fusion of the two feature types achieves best performance; (b) Ablation experiment of multi-source biological features in SMFM, showing percentage of variance of each ablation experiment; (c) illustrates performance of different feature selection methods, where multi-source feature selection can select feature set better than other feature selection methods; (d) compares the specific effects of gap values of PGKM features on the final performance; as the gap value increases, the performance increases.