SignalP 5.0 is an advanced tool for predicting signal peptides (SPs) in proteins, utilizing deep neural networks and Conditional Random Field (CRF) classification. This method improves upon previous versions of SignalP by enhancing prediction accuracy across all domains of life and distinguishing between three types of prokaryotic SPs: Sec/SPI, Sec/SPII, and Tat/SPI. The deep neural network architecture is better suited for recognizing sequence motifs of varying lengths, while the CRF imposes a defined grammar on the prediction, eliminating the need for post-processing. Transfer learning is also employed to improve performance on small datasets, particularly for archaeal sequences. Benchmarking against 18 SP prediction algorithms shows that SignalP 5.0 achieves high Matthews Correlation Coefficients (MCCs) for all types of SPs, with the highest MCCs for Tat/SPI predictions in Archaea and Gram-negative bacteria. The tool is available for both free academic use and commercial use after a fee.SignalP 5.0 is an advanced tool for predicting signal peptides (SPs) in proteins, utilizing deep neural networks and Conditional Random Field (CRF) classification. This method improves upon previous versions of SignalP by enhancing prediction accuracy across all domains of life and distinguishing between three types of prokaryotic SPs: Sec/SPI, Sec/SPII, and Tat/SPI. The deep neural network architecture is better suited for recognizing sequence motifs of varying lengths, while the CRF imposes a defined grammar on the prediction, eliminating the need for post-processing. Transfer learning is also employed to improve performance on small datasets, particularly for archaeal sequences. Benchmarking against 18 SP prediction algorithms shows that SignalP 5.0 achieves high Matthews Correlation Coefficients (MCCs) for all types of SPs, with the highest MCCs for Tat/SPI predictions in Archaea and Gram-negative bacteria. The tool is available for both free academic use and commercial use after a fee.