DEMO2: Assemble multi-domain protein structures by coupling analogous template alignments with deep-learning inter-domain restraint prediction

Abstract

Most proteins in nature contain multiple folding units (or domains). The revolutionary success of AlphaFold2 in single-domain structure prediction showed potential to extend deep-learning techniques for multi-domain structure modeling. This work presents a significantly improved method, DEMO2, which integrates analogous template structural alignments with deep-learning techniques for high-accuracy domain structure assembly. Starting from individual domain models, inter-domain spatial restraints are first predicted with deep residual convolutional networks, where full-length structure models are assembled using L-BFGS simulations under the guidance of a hybrid energy function combining deep-learning restraints and analogous multi-domain template alignments searched from the PDB. The output of DEMO2 contains deep-learning inter-domain restraints, top-ranked multi-domain structure templates, and up to five full-length structure models. DEMO2 was tested on a large-scale benchmark and the blind CASP14 experiment, where DEMO2 was shown to significantly outperform its predecessor and the state-of-the-art protein structure prediction methods. By integrating with new deep-learning techniques, DEMO2 should help fill the rapidly increasing gap between the improved ability of tertiary structure determination and the high demand for the high-quality multi-domain protein structures. The DEMO2 server is available at https://zhanggroup.org/DEMO/.

Graphical Abstract

DEMO2 is an significantly improved version for automated assembly of full-length structural models of multi-domain proteins by integrating analogous template alignments with deep-learning predicted inter-domain spatial restraints.

Figure 1.

Flowchart of the DEMO2 pipeline. The procedure mainly includes global and local templates identification, inter-domain spatial restraints prediction by DeepPotential, domain model assembly through fast L-BFGS simulation, and side-chain repacking and domain-domain linker reconstruction.

Figure 2.

Comparison of DEMO2 with AIDA and DEMO. (A) Head-to-head TM-score comparison of models assembled by DEMO2 and that created by DEMO. (B) Head-to-head TM-score comparison of models generated by DEMO2 and that built by AIDA. (C and D) representative examples are showing DEMO2 builds better full-length models than DEMO and AIDA. Gray and color cartoons are native structures and DEMO assembled models, respectively, and different domains in the assembled models are represented by different colors. (C) 1vz6A. (D) 4ewtA.

Figure 3.

Comparison of DEMO2 with DMPfold and trRosetta. (A) Violin plot plus box plot for the TM-score of the final full-length model, where IQR means the interquartile range of the TM-score. (B) Histogram of the rTM-score of the final full-length model, where the vertical line indicates the outlier of the TM-scores. (C and D) representative examples are showing DEMO2 creates more accurate models than DMPfold and trRosetta. Gray and color cartoons are native structures and DEMO2 assembled models, respectively, and different domains in the assembled models are represented by different colors. (C) 3arbA. (D) 1g87B.

Figure 4.

Example of the DEMO2 results page. (A) Title of the results page, link to download all results, FASTA sequence, and domain boundaries of the target. (B) The user provided domain models for the assembly. (C) Predicted residue-residue distance maps and contact maps for domain model assembly. (D) The top ten analogous templates identified by the analogous structural alignment. (E) Top five final full-length models assemble by the server and the estimated accuracy of the model, where different domains are represented by different colors.