CSI-LSTM

CSI-LSTM predicts protein secondary structure from backbone NMR chemical shifts using a bidirectional long short-term memory (biLSTM) neural network to map chemical shifts to secondary-structure states.

Key Features:

• BiLSTM Architecture: Employs a bidirectional long short-term memory (biLSTM) neural network to capture long-range dependencies in sequential chemical-shift data.
• NMR Chemical Shifts Input: Uses backbone chemical shifts from NMR spectroscopy, including hydrogen, carbon, and nitrogen nuclei, as model inputs.
• Mapping to Secondary Structure: Learns mappings from chemical shifts to secondary-structure states such as alpha-helices, beta-sheets, and random coils.
• Improved Accuracy: Reports enhanced accuracy relative to existing methods that use NMR chemical shifts, attributed to the biLSTM's ability to model complex patterns.

Scientific Applications:

• Protein Structure Analysis: Provides predicted secondary-structure elements to support interpretation of protein architecture.
• NMR Structural Studies: Complements NMR spectroscopy by translating chemical-shift data into secondary-structure assignments.
• Drug Design and Development: Supplies secondary-structure information relevant to assessing conformational landscapes for target characterization and ligand design.

Methodology:

Train a bidirectional LSTM on datasets of backbone NMR chemical shifts paired with known protein secondary-structure annotations so the model learns to map chemical shifts to secondary-structure states.