Deep learning based model for Cyro ET Sub-tomogram-Detection

High degree of structural complexity and practical imaging constraints make retrieval of macromolecular structures from cryo-ET is very challenging. For image classification of large-scale systematic macro-molecular structure from cryo-ET data.

For image classification of large-scale systematic macro-molecular structure from cryo-ET data, a deep learning-based image classification approach has been employed to improve the accuracy for a small range of SNR values where the present models have fallen short. Here, a novel SEC3 model for macro-molecule separation has been used.

The model comprises 3D convolutional blocks and 3D squeeze and excitation blocks. The model is trained on subtomogram datasets divided into 3 SNR values. Each SNR value namely- 0.03, 0.05, and infinity is further divided into 10 different classes based on their shape. The model understands by learning the valuable spatial information and spectral information available in the macromolecular structure data set. Valuable features detected by the convolution layers that are important for classification are given importance, thereby improving the overall model accuracy.

Data structure used for the project:

new data
|
|--SNR003
|    |
|    |--1bxn
|    |   |
|    |   |--subtomogram_mrc 
|    |   |--subtomogram_png
|    |   |--densitymap_mrc
|    |   |--densitymap_png
|    |   |--json_label
|    |   |--json_simulation
|    |
|    |
|    |--1f1b, 1yg6, 2byu, 2h12, 2ldb, 3gl1, 3hhb, 4d4r, 6t3e
|        |
|        |...(the same as 1bxn)
|    
|
|--SNR005
|    |
|    |...(the same as SNR003)
|  
|  
|--SNR infinity
     |
     |...(the same as SNR003)