Table 1. De novo phasing benchmark.

								Minimum F1 Å of an unambiguous Phaser solution†
Structure factors	Model sequence‡	Space group	No. of residues in model	No. of molecules in ASU	Solvent content (%)	No. of models, 100 CPU days§	No. of models, large-scale¶	Low-resolution models, 100 CPU days††	All-atom models, 100 CPU days††	All-atom models, large-scale††	Models, native constraints‡‡	Overall§§
1be7	1bq9	H3	51	1	43	3.5 × 105	1.7 × 107	—	—	—	0.882	0.882
1bq9	1bq9	P212121	51	1	43	3.5 × 105	1.7 × 107	—	—	—	0.627	0.627
2igd	1pgx	P212121	55	1	46	2.7 × 105	4.2 × 105	—	—	0.745	0.891	0.709
5cro	5cro	H32	55	4	70	2.6 × 105	7.4 × 105	—	—	0.927	0.982	0.709
1hz5	1hz6	P3221	61	2	72	2.3 × 105	7.3 × 105	—	0.541	0.656	0.787	0.541
1hz6	1hz6	P212121	61	3	59	2.3 × 105	7.3 × 105	—	0.672	0.689	0.836	0.639
1a32	1a32	P212121	65	1	41	2.8 × 105	2.8 × 105	—	0.754	0.708	0.800	0.677
1ctf	1ctf	P43212	68	1	47	2.4 × 105	3.2 × 105	—	—	—	0.882	0.515
1aar	1ubi	P1	71	2	35	2.0 × 105	5.4 × 107	—	—	—	—	1.000
1f9j	1ubi	I4122	71	2	60	2.0 × 105	5.4 × 107	—	—	—	—	0.901
1ubq	1ubi	P212121	71	1	33	2.0 × 105	5.4 × 107	—	—	0.690	0.662	0.549
2fcq	1ubi	P4332	71	2	58	2.0 × 105	5.4 × 107	—	—	—	—	0.915
2ojr	1ubi	P3221	71	1	73	2.0 × 105	5.4 × 107	—	—	—	0.549	0.549
1dt4	1dtj	P42212	74	1	54	2.8 × 105	4.9 × 105	0.649	0.622	0.500	0.635	0.419
1dtj	1dtj	C2	74	4	60	2.8 × 105	4.9 × 105	—	0.635	0.716	0.811	0.635
1ig5	1ig5	P43212	75	1	43	2.3 × 105	8.3 × 106	—	—	—	0.307	0.307
1cm3	1opd	P21	85	1	28	2.3 × 105	8.4 × 106	—	—	—	0.753	0.459
1opd	1opd	P1	85	1	33	2.3 × 105	8.4 × 106	—	—	—	0.800	0.800
1a19	1a19	I41	89	2	49	1.7 × 105	7.0 × 106	—	—	—	0.494	0.494
2hxx	1a19	C2	89	2	46	1.7 × 105	7.0 × 106	—	—	—	0.674	0.674
1mb1	1bm8	P41212	99	1	51	1.6 × 105	9.2 × 105	—	—	—	—	0.747
2hsh	1aiu	C2	105	1	35	1.5 × 105	4.4 × 105	—	—	0.400	0.600	0.400
1m6t	256b	C2221	106	1	43	1.8 × 105	1.5 × 105	—	0.453	0.443	0.491	0.283
256b	256b	P1	106	2	45	1.8 × 105	1.5 × 105	—	—	0.660	0.594	0.585
2bc5	256b	P212121	106	4	42	1.8 × 105	1.5 × 105	—	0.538	—	0.689	0.538
1elw	1elw	P41	117	2	47	1.5 × 105	1.1 × 105	—	0.453	0.521	0.897	0.436
1ab6	2chf	P31	128	2	57	1.2 × 105	3.5 × 106	—	—	0.508	0.398	0.398
2fka	2chf	F432	128	1	79	1.2 × 105	3.5 × 106	—	0.430	0.359	0.367	0.313
3chy	2chf	P212121	128	1	41	1.2 × 105	3.5 × 106	—	—	—	0.492	0.320
6chy	2chf	P212121	128	2	43	1.2 × 105	3.5 × 106	—	—	0.398	0.422	0.398

^† F _1 Å is a measure of model accuracy: the fraction of C^α atoms within 1 Å of the crystal structure of the modeled sequence. A dash (—) indicates that no models were found within the specified subset that gave an unambiguous Phaser solution.

^‡The Rosetta-modeled sequences were taken from an in-house curated benchmark used to test de novo modeling; in some cases the sequence does not include terminal segments (typically loops) or particular mutations present in the crystallized sequence.

^§Results of 100 CPU days per target without all-atom refinement, as is typically achievable by a state-of-the-art computer cluster; application of the same computational effort but including all-atom refinement led to pools of approximately one third the size.

^¶Results from 10⁴–10⁵ CPU days per target, with all-atom refinement, as is achievable with distributed computing.

^††Out of each pool of de novo models, the 200 models with best energies were tested for molecular replacement.

^‡‡Out of pools of approximately 50 000 models produced with the de novo method constrained with coarse native information for the backbone torsion angles, 40 models with the lowest C^α r.m.s.d. were tested for molecular replacement.

^§§Minimum F _1 Å that led to an unambiguous Phaser solution among all models tested in this study, including an additional 50 models with the lowest C^α r.m.s.d. to the crystal structure for each set (results not separately shown). These values are used as estimates of the ‘ease of phasing’ for each data set (see Table 2 ▶).