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

Problem and Solution on Calculating TM-score for protein/RNA/DNA Structures

TM-score is a metric for assessing the structural similarity of proteins and RNAs. Despite the simplicity in concept, there are numerous occasions when the TM-score was incorrectly calculated. In this page, we highlight the most frequent problem and design solutions to facilitate the calculation of TM-score.

(1) What is the most frequently seen problem of calculating TM-score between a predicted structure model and its native structure?

Figure 1. TM-score compares residues of the same order number which was read from Column 12-27 (red dashed box) of a PDB format or the _atom_site.auth_seq_id field of a PDBx/mmCIF format file. (A) The TM-score was correctly calculated because the residue numbers in two structures were correctly labeled and therefore correct residue pairs are read and compared (blue dotted lines). (B) The TM-score was incorrectly calculated because there is a shift on the order of residues of the native structure. Therefore, TM-score program matchs the incorrect residue pairs based on the order numbers read from the Column 12-17 of the PDB files (blue dotted lines).

(2) How to correctly calculate the TM-score between the model and native structures?

There are two general conditions:
• Condition 1: two structure files have consistent residue indexes (Figure 1A).
• Condition 2: two structure files have inconsistent residue indexes (Figure 1B).
For Condition-1, it is easy and you can simply run TMscore program using default setting. For Condition-2, you can run TMscore program with an option "-seq", in which the program first establishs the residue equivalence by sequence alignment and then calculate TM-score based on the obtained equivalence. Alternatively, you can use the option "-I align.txt", where align.txt is a pre-defined alignment file specifying the residue equivalence of two structures.


The following are the effective methods to calculate TM-score by different programs. They obtain the same TM-score results:

• Method 1 based on the TM-score program (for protein, RNA and DNA):

  $ TMscore model.pdb native.pdb				# Condition 1
  $ TMscore -seq model.pdb native.pdb			# Condition 2
  

• Method 2 based on the TM-align program (for protein):

  $ TMalign -seq model.pdb native.pdb			# Condition 2
  $ TMalign -I align.txt model.pdb native.pdb		# Condition 2
  

• Method 3 based on the RNA-align program (for RNA and DNA):

  $ RNAalign -TMscore 1 model.pdb native.pdb		# Condition 1
  $ RNAalign -seq model.pdb native.pdb			# Condition 2
  $ RNAalign -I align.txt model.pdb native.pdb		# Condition 2
  

• Method 4 based on the US-align program (for protein, RNA and DNA):

  $ USalign -TMscore 1 model.pdb native.pdb		# Condition 1
  $ USalign -seq model.pdb native.pdb			# Condition 2
  $ USalign -I align.txt model.pdb native.pdb		# Condition 2
  

(3) What is the difference between TM-score calculated by the above methods and TM-score calculated by the default option of TM-align, RNA-align, or US-align?

The above methods (1-4) compare two structures based on their given residue equivalency, i.e., based on the residue index in the PDB file or the given sequence alignment; this is called structure superposition. On the other hand, US-align, TM-align and RNA-align by default re-establish the optimal equivalent residues of two structures based on the structure similarity (regardless of the sequences) and then output a TM-score; this is called structural alignment.


• Method 5 based on the default option of US-align, RNA-align or TM-align:

  $ TMalign structure1.pdb structure2.pdb			# for protein
  $ RNAalign structure1.pdb structure2.pdb		# for RNA and DNA
  $ USalign structure1.pdb structure2.pdb			# for protein, RNA and DNA
  

The following figure shows the difference when comparing the same model to the same native structure, when using structure superposition (Methods 1-4) and structural alignment (Method 5):



(A)



$> TMscore model_1l6hA.pdb native_1l6hA.pdb
Number of residues in common=   69
RMSD of  the common residues=    5.986
TM-score    = 0.3486  (d0= 2.89)
(":" denotes the residue pairs of distance < 5.0 Angstrom)
AGCNAGQLTVCTGAIAGGARPTAACCSSLRAQQGCFCQFAKDPRYGRYVNSPNARKAVSSCGIALPTCH
   ::             ::::::::::::::::::::   ::::::: :  ::     :   :     
AGCNAGQLTVCTGAIAGGARPTAACCSSLRAQQGCFCQFAKDPRYGRYVNSPNARKAVSSCGIALPTCH

(B)



$> USalign model_1l6hA.pdb native_1l6hA.pdb
Aligned length=   65, RMSD=   0.49, Seq_ID=n_identical/n_aligned= 0.062
TM-score= 0.91722 (if normalized by length of Chain_1)
TM-score= 0.91722 (if normalized by length of Chain_2)
(":" denotes aligned residue pairs of d < 5.0 A)
----AGCNAGQLTVCTGAIAGGARPTAACCSSLRAQQGCFCQFAKDPRYGRYVNSPNARKAVSSCGIALPTCH
    :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::    
AGCNAGQLTVCTGAIAGGARPTAACCSSLRAQQGCFCQFAKDPRYGRYVNSPNARKAVSSCGIALPTCH----

Figure 2. An illustrative example for a comparison between the predicted and experimental structures of PDB 1l6h chain A. In this particular example, the structure was predicted incorrectly where the coordinates of all residues are shifted by 4 residues. (A) The direct superposition by TM-score program reflects the incorrect modeling and reports a low TM-score=0.349 and a high RMSD=5.99 A. (B) The structural alignments by US-align and TM-align programs try to perform a sequence-independent alignment based on structural similarity, reporting a much higher TM-score=0.917 and lower RMSD=0.49 A with a shifted sequence alignment between the model and native.

Back to the TM-score homepage