Table 1.

The averaged Pearson correlation coefficients (R_P) between the computed values (blind prediction for the topological features and regression for the rest of the models) and the experimental B-factors for a set of 364 proteins [65] and three sets of proteins of different sizes [70]. Top: Prediction R_Ps based on EH barcodes. Bottom: A comparison of the R_Ps of predictions from different methods based on the big protein set. Here, EH is the linear regression using EH^∞,0, EH^∞,1, EH^1,0, EH^1,1, EH^2,0, and EH^2,1 within each protein. For a few large and multi-chain proteins, to reduce the computation time and as a good approximation, we compute their EH barcodes on separated (protein) chains. The proteins that were analyzed on each separate chains include: 1F8R, 1H6V, 1KMM, 2D5W, 3HHP, 1QKI, and 2Q52 for both attractors; and additionally, 1GCO, 3LG3, 3W4Q, 2AH1, 3SZH, 4G6C for Rössler attractor. Note that there is an estimated upper limit (correlation coefficient of about 0.8) for B-factor prediction [75].

	All (364)		Small (33)		Medium (36)		Large (35)
Method	Lorenz	Rössler	Lorenz	Rössler	Lorenz	Rössler	Lorenz	Rössler
EH∞,0	0.586	0.469	0.476	0.504	0.569	0.531	0.565	0.500
EH∞,1	−0.039	0.119	−0.001	−0.010	−0.059	0.158	−0.062	0.105
EH∞,2	−0.097	0.003	−0.010	0.0	−0.099	0.0	−0.065	0.0
EH1,0	−0.477	0.486	−0.092	0.486	−0.521	0.542	−0.516	0.487
EH1,1	−0.381	0.204	−0.077	0.032	−0.384	0.276	−0.401	0.210
EH1,2	−0.104	0.002	−0.013	0.0	−0.105	0.0	−0.071	0.0
EH2,0	0.188	0.486	0.171	0.502	0.154	0.552	0.185	0.507
EH2,1	−0.258	0.015	−0.033	−0.022	−0.233	0.074	−0.276	−0.035
EH2,2	−0.100	0.002	−0.010	0.0	−0.102	0.0	−0.067	0.0
EH	0.691	0.698	0.746	0.773	0.701	0.729	0.663	0.665

Method	RP	Description
EH (Rössler)	0.698	Topological metrics
EH (Lorenz)	0.691	Topological metrics
mFRI	0.670	Multiscale FRI [65]
pfFRI	0.626	Parameter free FRI [64]
GNM	0.565	Gaussian network model [64]