DLEAMSE

DLEAMSE embeds mass spectra using deep learning to enhance spectral similarity scoring for proteomics, supporting improved protein identification and mass spectra clustering.

Key Features:

• Deep Learning Integration: Employs Siamese Networks trained on high-quality spectra from PRIDE Cluster to learn implicit spectral features for similarity scoring.
• Embedding Representation: Encodes spectra into compact 32-D vectors for downstream similarity calculations.
• Efficiency and Speed: Optimized for computational efficiency and GPU servers, achieving a significant reduction in computation time and performing similarity assessments at roughly one-third of the computation time required by the normalized dot product (NDP).
• Benchmarking and Performance: Demonstrates accuracy comparable to conventional spectral similarity scoring techniques while offering faster throughput.
• Persistent Embedding Storage: Allows storage of 32-D embeddings in a repository to enable rapid future comparisons without reprocessing raw spectra.
• Complementary Tool - mslookup: Includes mslookup for searching previously identified spectra in public repositories and spectral libraries to support in-house database creation.

Scientific Applications:

• Protein Identification: Improves matching of experimental spectra to theoretical or reference spectra to enhance protein identifications.
• Mass Spectra Clustering: Provides reliable similarity scores to improve clustering of related spectra in large-scale proteomics datasets.

Methodology:

The model is trained using Siamese Networks on spectra from PRIDE Cluster, encodes spectra into 32-D vectors, and computes similarity via Euclidean distance, with training and embedding optimized for GPU servers.