A machine learning approach for the identification of protein secondary structure elements from electron...

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Accurate identification of secondary structure elements (SSEs) from volumetric protein density maps is crucial for de novo backbone structure determination in electron cryo-microscopy (cryoEM).
• Automated and accurate detection of SSEs from medium-resolution (approximately 5-10 Å) cryoEM density maps remains a significant challenge.

Purpose of the Study:

• To present SSELearner, a novel machine learning approach for the automated identification of helices and beta-sheets in cryoEM density maps.
• To evaluate the performance of SSELearner using both simulated and experimentally derived cryoEM density maps.

Main Methods:

• Developed SSELearner, a machine learning model trained on existing volumetric maps from the Electron Microscopy Data Bank.
• Utilized SSID, a secondary structure annotator, to predict helices and beta-strands from the Cα trace.
• Tested SSELearner on 10 simulated and 13 experimentally derived cryoEM density maps.

Main Results:

• For simulated maps, SSELearner achieved high average specificity (94.9%) and sensitivity (95.8%) for helix detection, and 86.7% specificity and 96.4% sensitivity for beta-sheet detection.
• For experimental cryoEM maps, SSELearner demonstrated specificity of 91.8% and sensitivity of 74.5% for helix detection, and 85.2% specificity and 86.5% sensitivity for beta-sheet detection.
• Reduced accuracy on experimental maps highlights challenges compared to simulated data, yet indicates effectiveness of cross-map learning.

Conclusions:

• SSELearner provides an effective machine learning-based solution for automated secondary structure element identification in cryoEM density maps.
• The study demonstrates the feasibility and effectiveness of using one cryoEM map to train a model for detecting SSEs in another cryoEM map of similar quality.
• Further development is needed to address the challenges in SSE detection accuracy when dealing with experimentally derived cryoEM maps.