UniNovo

UniNovo performs de novo peptide sequencing from tandem mass spectrometry (MS/MS) spectra to reconstruct peptide sequences across CID, ETD, HCD, spectral pairs (e.g., CID/ETD) and non-tryptic digests.

Key Features:

• Universal de novo algorithm: A single algorithm intended to operate effectively across diverse MS/MS spectral types including CID, ETD, HCD and spectral pairs.
• Scoring function modeling ion dependencies: An innovative scoring function that captures dependencies between different ion types in spectra.
• Modified offset frequency learning: Ion-type dependency parameters are learned automatically using a modified offset frequency function.
• Probabilistic correctness estimation: Reports an estimated probability that each reconstructed peptide sequence is correct using simple statistical methods derived from a small training dataset.
• Benchmarking against established tools: Evaluated against PepNovo+, PEAKS, and pNovo with reported superior performance on ETD and competitive performance on CID and HCD.
• Protease diversity support: Demonstrated applicability to spectra from trypsin, LysC, and AspN digests.
• Implementation: Implemented in JAVA.

Scientific Applications:

• De novo peptide sequencing: Reconstruction of peptide sequences directly from MS/MS spectra without reliance on sequence databases.
• Cross-spectrum analysis: Analysis of different fragmentation methods (CID, ETD, HCD) and their combinations such as CID/ETD spectral pairs.
• Non-tryptic and alternative protease analyses: Sequencing of peptides generated by trypsin, LysC, AspN and other non-tryptic digests.
• Quality estimation of reconstructions: Assignment of confidence probabilities to peptide sequence reconstructions for downstream filtering or scoring.

Methodology:

Uses a universal de novo sequencing algorithm with a scoring function that models dependencies between ion types; learns those dependencies automatically via a modified offset frequency function; and estimates reconstruction correctness probabilities using simple statistical methods trained on a small dataset.