DiNovo enables high-coverage and high-confidence de novo peptide sequencing via mirror proteases and deep learning

Despite the recent advancements driven by deep learning, de novo peptide sequencing is still constrained by incomplete peptide fragmentation and insufficient protein digestion in current single protease-based proteomic experiments. Here, we present a software system, named DiNovo, for high-coverage and high-confidence de novo peptide sequencing by leveraging the complementarity of mirror proteases. DiNovo is empowered by several innovative algorithms, including a mirror-spectra recognition algorithm independent of pre-sequencing, two sequencing algorithms based on deep learning and graph theory, respectively, and target-decoy mapping, a method for sequencing result evaluation free of prior peptide identification. Compared with the trypsin protease used alone, DiNovo using two pairs of mirror proteases leads to two to three times high-confidence amino acids sequenced. Compared with previous single-protease de novo sequencing algorithms, DiNovo achieves much higher sequence coverage. DiNovo also shows great potential as a practical and powerful alternative to database search for peptide identification with quality control.