Published 2025 | Version v1

Multimodal deep learning integration of cryo-EM and AlphaFold3 for high-accuracy protein structure determination

  • 1. Department of Electrical Engineering and Computer Science, University of Missouri, Columbia, MO, 65211, United States
  • 2. NextGen Precision Health, University of Missouri, Columbia, MO, 65211, United States

Description

Cryo-EM is a key technology for determining the structures of proteins, particularly large protein complexes. However, automatically building high-accuracy protein structures from cryo-EM density maps remains a crucial challenge. In this work, we introduce MICA, a fully automatic and multimodal deep learning approach combining cryo-EM density maps with AlphaFold3-predicted structures at both input and output levels to improve cryo-EM protein structure modeling. It first uses a multi-task encoder-decoder architecture with a feature pyramid network to predict backbone atoms, Cα atoms and amino acid types from both cryo-EM maps and AlphaFold3-predicted structures, which are used to build an initial backbone model. This model is further refined using AlphaFold3-predicted structures and density maps to build final atomic structures. MICA significantly outperforms other state-of-the-art deep learning methods in terms of both modeling accuracy and completeness. Additionally, it builds high-accuracy structural models with an average template-based modeling score (TM-score) of 0.93 from recently released high-resolution cryo-EM density maps, showing it can be used for real-world, automated, accurate protein structure determination. 

Technical info

MICA Dataset Description

This dataset contains the complete materials for the MICA, including training data, pre-computed results, model checkpoints, and test datasets.

1. Results.tar.gz

Contains all pre-computed results generated by MICA, including processed outputs, predictions, and analysis results from the complete experimental pipeline.

2. Training_Dataset.tar.gz

Comprehensive training and validation dataset consisting of:

  • 550 cryo-EM density maps
  • Associated FASTA protein sequences
  • Ground truth PDB structures
  • AlphaFold3 structures docked against cryo-EM maps (combined into single PDB files)

This dataset was used for training and validating the MICA model.

3. Test_Dataset.tar.gz

Independent test dataset used for performance evaluation and benchmarking MICA against other state-of-the-art methods. 

4. trained_models.tar.gz

MICA's best-performing model checkpoint trained using PyTorch framework. This contains the optimized model weights and architecture parameters that achieved the highest performance on validation data.

5. input.tar.gz

Sample input data provided for quick testing and validation of MICA installation in conda environments. This lightweight dataset allows users to verify their setup and run initial tests without processing the full dataset.

Files

Files (67.8 GB)

Name Size
md5:02f3ac051fb0ab19e0767ffcababae2f
21.1 MB Download
md5:6df9fced135580bccba6a5fdbd4e0673
143.7 MB Download
md5:8157c382cdf5a09365fd193752bb49c8
20.2 GB Download
md5:a952451652294ca299ff5636f79524f7
154.7 MB Download
md5:7d85fc87250870cc25af1dd902fb3bc4
47.3 GB Download

Additional details

Funding

National Institutes of Health
R01GM146340

Software

Repository URL
https://github.com/jianlin-cheng/MICA
Programming language
Python