Deep learning guided mutagenesis of signal peptide in Escherichia coli

Signal peptides (SPs) play critical roles in protein secretion, yet their application in bioengineering remains limited by context-dependent functionality, host specificity, and genetic stability. This study presents a rational mutagenesis strategy for signal peptides in Escherichia coli under deep learning model and identifies the mutation-tolerant variants. At first, we generated 438 signal peptide sequences using SignalP 6.0. Through systematic modification of N-terminal, hydrophobic regions (H-region), and C-terminal sequence of signal peptide, we retrieved the high mutation tolerance existing in H-region. Among seven high mutation-tolerant signal peptides, SP22_54 from YjdP protein demonstrated exceptional robustness, maintaining functionality after extensive mutations. Further analysis revealed that functional SP mutants consistently preserved hydrophobic amino acid content with a minimal charge property. The SP_r1 mutant effectively mediated secretion of human carbonic anhydrase II (hCAII). The identified sequence determinants of mutation tolerance provide valuable insights for rational signal peptide design in biotechnological applications, enhancing reliability and cost-effectiveness of recombinant protein production across diverse expression systems.