Section: New Results

3D Convolutional Neural Networks for macromolecule localization in cryo-electron tomograms of intact cells

Participants : Emmanuel Moebel, Charles Kervrann.

In this study, we focus on macromolecule localization and classification in cryo-electron tomography (CET) images. Biologists are in need for efficient methods to localize macro-molecules (e.g. ribosomes) in frozen cell samples. The high amount of noise and imaging artifacts are the reasons why very few computational methods exist for this task. In fact, the most used method today is template matching (TM) whose resulting score map comprises a high amount of false positives. Therefore, it is necessary to apply post-processing techniques (ROI selection, classification) in order to refine the localization results. We propose an alternative localization method to TM, based on a convolutional neural network (CNN). The idea is to propose a robust and more straight-forward approach, allowing to bypass the conventional processing chain. By using python toolboxes optimized for GPU computing (elektronn, keras), we are able to reach computation time much lower than the current approach. Results on synthetic data demonstrate the superiority of our approach compared to TM. In addition, we applied our method on experimental data in order to localize sub-classes of ribosomes (membrane-bound and cytoplasmic ribosomes), a task difficult to achieve with TM alone. We are currently in the process of publishing these results. Future perspectives include localizing smaller macro-molecules, like proteasomes.

Collaborators: Damien Larivière (Fondation Fourmentin-Guilbert),

                          Julio Ortiz, Antonio Martinez (Max-Planck Institute, Martinsried, Germany).

Figure 10. Illustration of 3D CNN to localize ribosomes isolated in the cytoplasm and close to the cell membrane in cryo-electron tomography (courtesy of Max-Planck Institute, Martinsried, Germany).