Identification of Macromolecules in Cryo-Electron Microscopy Reconstruction Maps Using Neuronal Networks

 

 

 

 

 

Identification of Macromolecules in Cryo-Electron Microscopy Reconstruction Maps Using Neuronal Networks

Supervisors: Prof. Arwen Pearson (UHH), Dr. Andrea Thorn (UHH), Prof. Nihat Ay (TUHH)

Single-particle cryo electron microscopy (Cryo-EM) is developing incredibly quickly, providing unprecedented structural information about large molecular machines, membrane complexes, and especially interaction between proteins. This rapid progress has been facilitated by game-changing technical advances in sample preparation and instrumentation. However, the computational methods needed to analyse the data care lagging well behind what is experimentally possible. Map interpretation, for example, can be challenging, in particular if the constituting structures require de-novo model building or are very mobile. Convolutional neural networks combine traditional image analysis with machine learning by cascading layers of trainable convolution filters and are exceptionally well suited for map annotation.
We developed a network, Haruspex, to annotate secondary structures and oligonucleotides in Cryo-EM reconstructions at high resolutions. Our annotated data, generated by a preprocessing pipeline and manual curation, constitute the largest data set used in Cryo-EM map interpretation by neural networks so far. HARUSPEX has recently become available as part of CCP-EM. Now, we would like to build on this proof-of-principle and adapt the method for more structural features, lower map resolutions and investigate how it could be exploited for automatic model building. We also aim to better understand how Haruspex recognizes secondary structures in 3D maps.

Apply here!