Home Research COVID-19 Services Publications People Teaching Job Opening News Forum Lab Only
Online Services

I-TASSER I-TASSER-MTD C-I-TASSER CR-I-TASSER QUARK C-QUARK LOMETS MUSTER CEthreader SEGMER DeepFold DeepFoldRNA FoldDesign COFACTOR COACH MetaGO TripletGO IonCom FG-MD ModRefiner REMO DEMO DEMO-EM SPRING COTH Threpp PEPPI BSpred ANGLOR EDock BSP-SLIM SAXSTER FUpred ThreaDom ThreaDomEx EvoDesign BindProf BindProfX SSIPe GPCR-I-TASSER MAGELLAN ResQ STRUM DAMpred

TM-score TM-align US-align MM-align RNA-align NW-align LS-align EDTSurf MVP MVP-Fit SPICKER HAAD PSSpred 3DRobot MR-REX I-TASSER-MR SVMSEQ NeBcon ResPRE TripletRes DeepPotential WDL-RF ATPbind DockRMSD DeepMSA FASPR EM-Refiner GPU-I-TASSER

BioLiP E. coli GLASS GPCR-HGmod GPCR-RD GPCR-EXP Tara-3D TM-fold DECOYS POTENTIAL RW/RWplus EvoEF HPSF THE-DB ADDRESS Alpaca-Antibody CASP7 CASP8 CASP9 CASP10 CASP11 CASP12 CASP13 CASP14

D-I-TASSER is a composite approach to protein structure prediction extended from I-TASSER for high-accuracy protein structure and function prediction. D-I-TASSER is proposed to construct full-length protein structure models by integrating multiple deep-learning restraints with cutting-edge iterative threading assembly simulations. On our benchmark set with proteins lacking homologous templates, D-I-TASSER generates high-quality models with average TM-score 94% higher than I-TASSER, in terms of the structure model TM-scores.

Methods

D-I-TASSER is a hierarchical pipeline for hybrid deep-learning and threading-assembly based protein structure prediction. Compared to its predecessor I-TASSER, the major new development is the construction and integration of deep-learning spatial restraints with the I-TASSER-based threading-fragment assembly simulations. As depicted in Figure 1, D-I-TASSER starts with the generation of multiple sequence alignment (MSA) from DeepMSA, where a deep residual convolutional network architecture (named AttentionPotential) is then extended to create spatial restraint predictions on contact and distance maps and hydrogen-bonding networks. Meanwhile, LOMETS3 is used to identify template structures and alignments by integrating multiple contact- and profile-based threading algorithms. Guided with the deep-learning and threading restraints and the inherent I-TASSER force field, D-I-TASSER folding simulations generate multiple decoy conformations through Replica Exchange Monte Carlo simulations. The lowest free energy states are then identified by decoy structure clustering through SPICKER. The final atomic models are refined by fragment-guided molecular dynamics (FG-MD) simulations with sidechains repacked with FASPR. The quality of the D-I-TASSER model is assessed by the estimated TM-score (eTM-score).


Figure 1. Pipeline of D-I-TASSER.


Performance of D-I-TASSER server

CASP (or Critical Assessment of Techniques for Protein Structure Prediction) is a community-wide experiment for testing the state-of-the-art of protein structure prediction, which has taken place every two years since 1994. The experiment is strictly blind because the structures of the test proteins are unknown to the predictors. The D-I-TASSER server (as "Zhang-Server") participated in the Server Section of CASP14 (2020), and was ranked as the No 1 automated server (Figure 2).


Figure 2. The results of the automated Server Section in the 14th Critical Assessment of protein Structure Prediction (CASP14) experiment. The top 10 protein structure prediction servers are shown in descending order of the Z-score of the GDT_TS score, which is a metric used to quantify the overall structure quality. The top two servers were the D-I-TASSER server (named ‘Zhang-Server’ in CASP14), and the C-QUARK server (named ‘QUARK’ in CASP14), both from our group. This figure is redrawed based on the data from CASP14 website.




Server inputs

The user needs to paste the fasta-formatted amino acid sequence into the input box, or upload the amino acid sequence of the query protein using the "Choose file" button.


Figure 3. Input of D-I-TASSER.


Server outputs
The output of the D-I-TASSER server includes: An illustrative example of the D-I-TASSER output can be seen from below:
  • Secondary structure, solvent accessibility, contact map,distance Map, and hydrogen bond networks information:

    Figure 4. Secondary structure, solvent accessibility, contact map,distance Map, and hydrogen bond networks information in the D-I-TASSER output.

  • Templates, final models, and analog information:

    Figure 5. Templates, final models, and analog information in D-I-TASSER output.

  • Gene Ontology (GO) Term prediction information:

    Figure 6. Gene Ontology (GO) Term prediction information in the D-I-TASSER output. This information is output only after you check the "Predict protein function based on structure model (running time may be doubled)." option of the input.

  • Enzyme Commission (EC) and ligand binding site prediction information:

    Figure 7. Enzyme Commission (EC) and ligand binding site prediction information in the D-I-TASSER output. This information outputs only after you check the "Predict protein function based on structure model (running time may be doubled)." option of the input.


    Output tips
    The output of the D-I-TASSER modeling results are generally summarized in a webpage, the link of which is sent to the user by their registered email after the modeling is completed. In the following, we present answers to several most frequently asked questions in interpreting the D-I-TASSER results:

    • What are the 'top 10 threading templates used by D-I-TASSER'?

      D-I-TASSER modeling starts from the structure templates identified by LOMETS3 from the PDB library. LOMETS3 is a meta-server threading approach containing multiple threading programs, where each program can generate tens of thousands of templates. D-I-TASSER only uses the templates of the highest significance in the threading alignments, which are measured by the Z-score (the difference between the raw and average scores in the unit of standard deviation). The top 10 templates are the 10 templates selected from the LOMETS3 threading programs. Usually, one (or two) template with the highest Z-score is selected from each threading program, where the threading programs are sorted by the average performance in the large-scale benchmark test experiments.

    • What are the 'top final models from D-I-TASSER'?

      For each target, D-I-TASSER simulations generate tens of thousands of conformations (called decoys). To select the final models, D-I-TASSER uses the SPICKER program to cluster all the decoys based on pair-wise structure similarity, and report up to five models which correspond to the five largest structure clusters. In Monte Carlo theory, the largest clusters correspond to the states of the largest partition function (or lowest free energy) and therefore have the highest confidence. The confidence of each model is quantitatively measured by eTM-score (see below). Since the top 5 models are ranked by the cluster size, it is possible that the lower-rank models have a higher eTM-score. Although the first model has a higher eTM-score and a better quality in most cases, it is not unusual that the lower-rank models have a better quality than the higher-rank models. If the D-I-TASSER simulations converge, it is possible to have less than 5 clusters generated. This is usually an indication that the models are high quality because of the converged simulations.

    • What are 'Proteins with similar structure'?

      After the structure-assembly simulation, D-I-TASSER uses the TM-align program to match the first D-I-TASSER model to all structures in the PDB library. This section reports the top 10 proteins from the PDB which have the closest structural similarity (i.e. the highest TM-score) to the predicted D-I-TASSER model. Due to their structural similarity, these proteins often have similar function to the target. However, users are encouraged to use the function prediction in D-I-TASSER output to obtain the biological function of the target protein, since D-I-TASSER predicts the function using COACH and COFACTOR, which have been extensively trained to derive function from many sequence and structure features, and as a result, these programs have a much higher accuracy than function annotations derived only from the global structure comparison.

    • How can I know if my model is successfully folded?

      Since the experimental structures are unknown for the user input sequence, we have designed an estimated TM-score (eTM-score) to quantitatively estimate the quality of the D-I-TASSER models. The eTM-score is a linear combination of three components: significance of the LOMETS3 threading alignments, satisfaction rate of the predicted contact-maps, the model fitting rate of predicted distance-maps, and the decoy convergence degree of the D-I-TASSER simulations. Based on benchmark testing, the eTM-score had a Pearson correlation coefficient (PCC) of 0.757 with TM-score. As a result of this high correlation, we were able to select a eTM-score cutoff of 0.5, corresponding to an estimated TM-score=0.5, and attain a Matthews correlation coefficient (MCC) on the benchmark dataset of 0.644 and a false discovery rate (FDR) of only 2.71%. Therefore, the D-I-TASSER models with eTM-score > 0.5 are considered to be successfully folded.

    • What is eTM-score?

      eTM-score is designed to quantitatively evaluate the quality of the D-I-TASSER models. It is derived from a linear combination of 4 components, including the significance of LOMETS threading alignments, the satisfaction rate of predicted contact-maps, the model fitting rate of predicted distance-maps, and the decoy convergence degree of D-I-TASSER simulations. A eTM-score of higher value signifies a model of high confidence.

    • What is TM-score?

      TM-score is a metric for measuring the structural similarity between two structures (see Zhang and Skolnick, Scoring function for automated assessment of protein structure template quality, Proteins, 2004 57: 702-710). The purpose of proposing TM-score is to solve the problem of RMSD which is sensitive to local errors. Because RMSD is an average distance of all residue pairs in two structures, a local error (e.g. a misorientation of the tail) will result in a big RMSD value although the global topology is correct. In TM-score, however, the small distance is weighted stronger than the big distance, which makes the score insensitive to local modeling errors. A TM-score > 0.5 indicates a model of correct topology and a TM-score < 0.17 means a random similarity. These cutoffs are not dependent on the protein length.

    • What is difference and relationship between eTM-score and TM-score?

      TM-score (or RMSD) is a known standard for measuring structural similarity between two structures and is typically used to measure the accuracy of structure modeling when the native structure is known. eTM-score is a metric that was developed for D-I-TASSER to estimate the confidence of modeling. In the case where the native structure is not known, it becomes necessary to use the eTM-score predict the quality of the modeling prediction, i.e. the distance between the predicted model and the native structures.

    • In a benchmark test set of 797 proteins, we found that eTM-score is highly correlated with TM-score. The correlation coefficient of the eTM-score of the first model with the TM-score to the native structure is 0.757. These data lay the base for the reliable prediction of the TM-score using eTM-score. In the output section, D-I-TASSER reports the eTM-scores of all predicted models for reference.



    How to cite D-I-TASSER?
    • Wei Zheng, Yang Li, Qiqige Wuyun, Xiaogen Zhou, Chengxin Zhang, Robin Pearce, Eric W. Bell, Yiheng Zhu, Yang Zhang. Integrating deep neural network models with I-TASSER for accurate protein structure prediction. To Be Submitted. (2021) [PDF] [Support Information]

    [back to server]

  • zhanglabzhanggroup.org | +65-6601-1241 | Computing 1, 13 Computing Drive, Singapore 117417