The partitioning of the structure into domains may result in domains consisting of contiguous stretches of polypeptide chain – one stretch per domain (contiguous domains). Frequently, however, regions of the polypeptide chain that are distant in sequence, are close together in 3D structure, thus a domain may consist of two or more segments of the chain which are non-contiguous in sequence (non-contiguous domains). We note that the average fragmentation of domains correlates with the average number of domains assigned by each method: if an automatic method assigns on average more domains it also assigns on average more fragments per domain. The fragmentation of domains is presented from three perspectives: a fraction of correctly/incorrectly fragmented domains, partitioning of domains into discontinuous fragments by each method, and correlation between partitioning protein into domains and the fragmentation of domains.

1. Fraction of correct/incorrect fragmented domains

Tabular data for the figure

Number of fragments/per domain	Expert consensus	PDP	DomainParser	PUU	NCBI
1	91.7	82.4	91.3	72.2	85
2	7.6	14.9	8	18.1	13.1
3	0.7	2.7	0.7	7.2	1.9
4+	0	0	0	2	0

Evaluation of domain assignment methods using using fragmentation of domains. The evaluation is performed using Balanced_Domain_Benchmark_2.

2. Partitioning of domains into discontinuous fragments by each method

Tabular data for the figure

Number of fragments	PDP	DomainParser	PUU	NCBI
1 fragment	-9.3	-0.4	-19	-6.7
2 fragments	7.3	0.4	10.5	5.5
3 fragments	2	0	6.5	1.2
4+ fragments	0	0	2	0

Fraction of continuous and discontinuous domains mis-assigned by each method and binned by number of fragments per domain. The evaluation is performed using Balanced_Domain_Benchmark_2.

3. Comparing tendency toward correct assignment on the level of protein vs. the level of domain

Tabular data for the figure

	PDP	DomainParser	PUU	NCBI	Expert consensus
average # of domains per chain	2.03	1.78	2.11	2.03	1.95
average # of fragments per chain	1.20	1.09	1.49	1.30	1.09

Side-by-side comparison of the average number of fragments per domain and average number of domains per chain as assigned by each method. The left Y-axis scale refers to the average number of fragments per domain and the right Y-axis scale refers to the average number of domains per chain. The average number of fragments per domain is calculated using Y / X, where Y is the sum of all fragments assigned for each domain and X is the total number of domains assigned. The average number of domains per chain is calculated using A / B, where A is the sum of all domains assigned for each chain and B is the total number of chains. The proportion between average fragments and average number of domains is 1:1.65. The evaluation is performed using Balanced_Domain_Benchmark_2.