D-I-TASSER (Deep learning-based Iterative Threading ASSEmbly Refinement) is a new method extended from I-TASSER
for deep learning-based, high-accuracy protein structure and function predictions.
Starting from a query sequence, D-I-TASSER first creates multiple sequence alignments (MSAs)
by DeepMSA2 via iteratively searching of genomics and metagenomics sequence
databases, where inter-residue contact/distance maps and hydrogen-bond (HB) networks are generated
by three complementary deep neural-network predictors from DeepPotential,
AttentionPotential, and AlphaFold2 (optional in 'Advanced options').
Meanwhile, multiple template alignments are identified from the PDB by the DeepMSA2-guided meta-threading
program LOMETS3.
The full-length structural models are finally constructed by iterative fragment assembly Monte Carlo
simultions under the guidance of the I-TASSER force field and deep-learning
contact/distance/HB restraints, where a new domain spliting and reassemly module is introduced
for modelling large-size multi-domain proteins.
Finally, the biological functions of the query protein are derived using the structure-based function
annotation method COFACTOR.
The D-I-TASSER pipeline (as 'UM-TBM') ranked as the No. 1 server in both
Single-domain and
Multi-domain sections in the CASP15 experiment. Notably, D-I-TASSER achieves higher accuracy than both AlphaFold2 and AlphaFold3 in
recent CASP experiments and large-scale benchmark evaluations.
The server is freely accessible to all users, including commercial ones.
Please contact Wei Zheng for any problems or questions.
To model Multi-chain target, please use our protein complex structure prediction server, DMFold.