AI Driven Structure Modeling & Validation

AI-powered software for structure modeling and validation for cryo-EM maps. Build protein structures with DeepMainmast, model DNA and RNA with CryoREAD, and evaluate local model quality with DAQ-Score.

DeepMainmast: From Cryo-EM to Atomic Model

DeepMainmast is an automated software platform for protein structure modeling from cryo-EM maps. It uses a cryo-EM map and the corresponding protein sequences as input and generates structural models for single proteins and multi-chain protein complexes.

DeepMainmast uses deep learning to detect amino-acid and backbone-atom features in cryo-EM maps. It traces Cα backbone paths, assigns protein sequences, and assembles compatible fragments into complete models. It can also incorporate structure models generated by AI tools such as AlphaFold, using regions that agree with the experimental density to improve difficult or low-resolution areas.

DeepMainmast was published in Nature Methods in 2024:

Terashi, G., Wang, X., Prasad, D., Nakamura, T., & Kihara, D. (2024). DeepMainmast: integrated protocol of protein structure modeling for cryo-EM with deep learning and structure prediction. Nature Methods, 21(1), 122-131.

Contact

CryoREAD: Build DNA and RNA Structures from Cryo-EM Maps

CryoREAD is fully automated software for de novo DNA and RNA structure modeling from cryo-EM maps. It takes a cryo-EM density map and nucleic-acid sequences as input and generates atomic models of DNA and RNA.

CryoREAD uses deep learning to detect nucleic-acid density and identify local structural features, including phosphate, sugar, and base positions. It then traces the nucleic-acid backbone, assigns the target sequence, and assembles compatible fragments into complete atomic models. This automated workflow is particularly valuable for maps where DNA or RNA structure modeling is difficult because the resolution is coarser than the atomic level.

CryoREAD was published in Nature Methods in 2023:

Wang, X., Terashi, G., & Kihara, D. (2023). CryoREAD: de novo structure modeling for nucleic acids in cryo-EM maps using deep learning. Nature Methods, 20, 1739–1747.

Contact

DAQ-Score: Validate Protein Structure Models in Cryo-EM Maps

Protein structure models built from cryo-EM maps can contain local modeling errors that are difficult to identify using conventional validation methods. These include incorrect amino-acid assignments and residue shifts.

DAQ-Score uses deep learning to evaluate the compatibility between each residue in a protein model and the local cryo-EM density. By highlighting potentially incorrect regions, DAQ-Score helps researchers identify and correct modeling errors, supporting the construction of more accurate and reliable protein structure models.

DAQ-Score was published in Nature Methods in 2022:

Terashi, G., Wang, X., Maddhuri Venkata Subramaniya, S. R., Tesmer, J. J. G., & Kihara, D. (2022). Residue-wise local quality estimation for protein models from cryo-EM maps. Nature Methods, 19, 1116–1125.

Contact