Fig. 3: Overview of the AMP screening approach.
a, Ensembles of NRSPDs. The window size is set to L, with a step size of L/2, where L ranges from 8 to 30. A total of 5 NRSPDs were constructed through a sliding-window technique employed on the complete UniProtKB/Swiss-Prot database, encompassing 410,192,277 short peptide sequences. b, The whole process of SPEL includes modules of AMPSorter, BioToxiPept and wet-lab validation. The antimicrobial activities of the output peptide sequences were validated by wet-lab experiments. c, Ensembles of GNRSPDs. The sequences generated by AMPGenix at the default temperature = 1 parameter were used to construct 10 different GNRSPDs, comprising a total of 7,798 unique short peptide sequences. d,e, Length distribution for 154 m_AMPs (d) and 42 g_AMPs (e). f–h, The prediction probabilities of AMPSorter (f), BioToxiPept (h) and RS predicted by QSAR (g) of 154 m_AMPs and 42 g_AMPs. The boxplots centre on the median and extend to the 25th and 75th percentiles, and the whiskers extend to the furthest point within 1.5× the interquartile range. A two-sided Mann–Whitney U-test was performed for comparison. Exact P values are provided in Source Data Fig. 3. i, UMAP visualization of AMPs, m_AMPs and g_AMPs using k-mer encoding (k = 3). Each point represents a sequence, with the position determined by reducing the high-dimensional k-mer feature space into two dimensions. The schematic in b was created with BioRender.com.
