Fig. 3: D-I-TASSER modeling results on 230 multidomain proteins.
From: Deep-learning-based single-domain and multidomain protein structure prediction with D-I-TASSER
a,b, Head-to-head TM score comparisons between the D-I-TASSER and AlphaFold2 on full-chain modeling (a) and domain-level modeling (b). c, TM score comparison of D-I-TASSER and AlphaFold2 on two-domain, three-domain and high-order domain proteins. The height of the histogram indicates the mean value and the error bar depicts s.d. d, D-I-TASSER and AlphaFold2 models for C. reinhardtii flagellar radial spoke protein (PDB ID: 7jtkB) superposed with the target structure, where two domains of the target structure are colored differently. e, The residue–residue distance map (heat map) along with the number of aligned residues per site (n, shown in margins) predicted from AlphaFold2 (upper triangle) versus that calculated from the target structure (lower triangle) for PDB ID: 7jtkB. f, As in e, but modeled with D-I-TASSER. g, D-I-TASSER and AlphaFold2 models for human InaD-like protein (PDB ID: 6irdC) superposed with the target structure, where two domains of the target structure are colored differently. h,i, Equivalent to e,f, respectively, but for PDB ID: 6irdC.
