Signal peptides play an important role in guiding and transferring transmembrane proteins and secreted proteins. In recent years, with the explosive growth of protein sequences, computationally predicting signal peptides and their cleavage sites from protein sequences is highly desired.

In this work, we present an improved approach, Signal-3L 3.0, for signal peptide recognition and cleavage-site prediction using a 3-layer hybrid method of integrating deep learning algorithms and domain rules. There are three main components in Signal-3L 3.0 prediction engine: 

(1) a deep bidirectional long short-term memory (biLSTM) network with a soft attention mechanism learns abstract features from sequence to judge whether a query protein containing signal peptide;

(2) the statistics propensity rule-based cleavage site screening method is applied to generate the set of candidate cleavage sites;

(3) a conditional random field with a hybrid convolutional neural network (CNN) and biLSTM is trained for predicting the final unique cleavage site. Experimental results on the benchmark datasets show that the new deep learning-driven Signal-3L 3.0 yields promising performance.

Fig. 1. The overall prediction protocol of Signal-3L 3.0.

Fig. 2. The architecture of the sequence-level classifier for predicting whether a protein sequence contains signal peptide (SP) ,transmembrane helix(TM) or belongs to globular proteins (N/C).

Fig. 3. The architecture of the residue-level classifier for predicting the potential cleavage site.