Haruspex: A Neural Network for the Automatic Identification of Oligonucleotides and Protein Secondary Structure in Cryo‐Electron Microscopy Maps

Abstract

In recent years, three‐dimensional density maps reconstructed from single particle images obtained by electron cryo‐microscopy (cryo‐EM) have reached unprecedented resolution. However, map interpretation can be challenging, in particular if the constituting structures require de‐novo model building or are very mobile. Herein, we demonstrate the potential of convolutional neural networks for the annotation of cryo‐EM maps: our network Haruspex has been trained on a carefully curated set of 293 experimentally derived reconstruction maps to automatically annotate RNA/DNA as well as protein secondary structure elements. It can be straightforwardly applied to newly reconstructed maps in order to support domain placement or as a starting point for main‐chain placement. Due to its high recall and precision rates of 95.1 % and 80.3 %, respectively, on an independent test set of 122 maps, it can also be used for validation during model building. The trained network will be available as part of the CCP‐EM suite.

Cryo2StructData: A Large Labeled Cryo-EM Density Map Dataset for AI-based Modeling of Protein Structures

“…Both DeepTracer and ModelAngelo automatically generate the atomic models of the protein, but they have not released their processed data. Haruspex 20 used 293 experimental maps for training and an independent test set of 122 experimental maps for testing, and EMNUSS 21 used 120 and 43 experimental density maps for training and testing respectively. CR-I-TASSER 22 , EMNUSS 21 , and Emap2sec 23 employed a hybrid approach that combined both simulated maps and experimental maps in their training and validation processes.…”

“…The dimensions of density maps of different proteins vary and are usually too large to fit into the memory of a standard GPU for training deep learning models. Similar to the approach employed in DeepTracer 18 , Haruspex 20 , CR-I-TASSER 22 , Emap2sec 23 , and Cascaded-CNN 13 , we performed grid division to divide the density maps into 3D subgrids with dimension of 32 × 32 × 32 overlapped by 6 voxels on each face of the subgrid to train deep learning methods. We choose the dimension 32 as it is big enough to capture the patterns (e.g.…”

Integrating model simulation tools and cryo‐electron microscopy

“…Here, a single pass of the cryo‐EM map (1) through the network is sufficient to identify all SSEs in the map and DNA (2, α‐helices in red, b‐sheets in blue, DNA in orange). Example adapted from Reference 79.…”

“…Haruspex 79 detects Oligonucleotides along with SSEs in cryo‐EM density maps (Figure 4c). The prediction model is trained on 293 experimentally derived EMDB maps with an average resolution of 4 Å or better and tested on a test set of 122 EMDB maps with a resolution of 4.4 Å or better.…”

Identification of RNA Structure Over the Protein Surface Using Neural Network

“…The emblem image for every RBP or analytic ratings on specific RNA sequences may be used to depict specific trends acquired using these approaches is generally [8].Interacting processes for several of these described RBPs, including hnRNP, Nova, and PAZ, have also been elucidated using structural deconstruction approaches. Despite these accomplishments, experimental tests are limited by reaction, detecting, and scaling constraints [9].Although pyrimidines were higher photo activatable than purines, Ultraviolet crosslinking tests preferred uridine-rich patterns. However, ribonucleoprotein co-crystals could plausibly confirm the biochemical basis of the tested particularities, one and a several like these co-crystals would barely describe the confusing patterns on emblem diagrams [10].Computational techniques can help improve experimental outcomes in that regard.…”