Towards accurate modeling of noncovalent interactions for protein rigidity analysis

BMC Bioinformatics

Table 2 Example B-cubed scoring calculations.

	D1		D2		D3
Residue	Re	Pr	Re	Pr	Re	Pr
1	$\frac{5}{5}$	$\frac{5}{8}$	$\frac{5}{5}$	$\frac{5}{10}$	$\frac{1}{5}$	$\frac{1}{1}$
2	$\frac{5}{5}$	$\frac{5}{8}$	$\frac{5}{5}$	$\frac{5}{10}$	$\frac{1}{5}$	$\frac{1}{1}$
3	$\frac{5}{8}$	$\frac{5}{5}$	$\frac{5}{10}$	$\frac{1}{5}$	$\frac{1}{1}$	$\frac{5}{5}$
4	$\frac{5}{5}$	$\frac{5}{8}$	$\frac{5}{5}$	$\frac{5}{10}$	$\frac{1}{5}$	$\frac{1}{1}$
5	$\frac{5}{5}$	$\frac{5}{8}$	$\frac{5}{5}$	$\frac{5}{10}$	$\frac{1}{5}$	$\frac{1}{1}$
6	$\frac{2}{2}$	$\frac{2}{8}$	$\frac{2}{2}$	$\frac{2}{10}$	$\frac{1}{2}$	$\frac{1}{1}$
7	$\frac{2}{2}$	$\frac{2}{8}$	$\frac{2}{2}$	$\frac{2}{10}$	$\frac{1}{2}$	$\frac{1}{1}$
8	$\frac{1}{3}$	$\frac{1}{8}$	$\frac{3}{3}$	$\frac{3}{10}$	$\frac{1}{3}$	$\frac{1}{1}$
9	$\frac{2}{2}$	$\frac{2}{3}$	$\frac{3}{3}$	$\frac{3}{10}$	$\frac{1}{3}$	$\frac{1}{1}$
10	$\frac{2}{2}$	$\frac{2}{3}$	$\frac{3}{3}$	$\frac{3}{10}$	$\frac{1}{3}$	$\frac{1}{1}$
Avg	0.93	0.51	1.0	0.38	0.30	1.0
F1	0.65		0.55		0.46

Shown are the calculations of B-cubed precision, recall, and F1-scores for the small examples shown in Figure 7. The scoring method is described in the Methods section. The B-cubed score (shown as F1) for decomposition D1 is higher than the all-rigid (D2) and all-floppy (D3) decompositions. In the all-rigid baseline, all 10 residues are placed into the same cluster (D2), resulting in 100% recall but low precision. To contrast, in the all-floppy baseline, each residue is placed in a unique cluster resulting in 100% precision but low recall.

ISSN: 1471-2105