Image-centric compression of protein structures improves space savings

BMC Bioinformatics

Table 2 Compression of entire PDB/mmCIF files benchmark

	PDB/	BCIF	MMTF	MMTF	PIC (lossy precision)				Foldcomp
Protein	CIF size	Size	Size	C\(\alpha\)-lossy	Coord	MMTF-	Total	RMSD	(Lossy precision)
	(KB)	(KB)	(KB)	(KB)	(KB)	Meta (KB)	(KB)	(Å)	(KB)
2ja9	163	24	13	3	9	4	14	0.030680	segfault
2jan	1101	108	98	15	61	22	84	0.047302	26
2jbp	2397	224	214	30	108	52	161	0.043140	runtime>day
2ja8	2831	299	249	42	138	59	197	0.043373	runtime>day
2ign	3579	329	321	41	147	71	219	0.068555	runtime>day
2jd8	4457	382	384	49	196	88	285	0.056100	runtime>day
2ja7	5605	534	488	74	258	110	369	0.055422	runtime>day
2fug	6386	580	566	82	283	128	412	0.060134	runtime>day
2b9v	6817	594	608	73	288	128	417	0.072982	runtime>day
2j28	8152	738	586	70	346	50	396	0.055343	runtime>day
6hif	12726	877	894	167	372	194	566	0.061988	runtime>day
3j7q	16027	1211	1027	108	475	234	710	0.058150	runtime>day
3j9m	17995	1339	1198	160	525	282	807	0.069300	runtime>day
6gaw	20825	1584	1359	184	587	324	911	0.070872	runtime>day
5t2a	22787	1628	1396	151	651	262	914	0.068460	segfault
4ug0	24906	1827	1606	177	707	364	1072	0.069050	runtime>day
4v60	24377	1509	1618	242	730	191	922	0.119788	491
4wro	35661	2336	1902	171	848	447	1296	0.085947	runtime>day
6fxc	31328	1961	1678	181	917	103	1020	0.099960	segfault
4wq1	40130	2646	2212	216	968	523	1492	0.086623	runtime>day

Actual compressed file sizes. We compare PIC, MMTF, Foldcomp, and BCIF formats. However, because PIC does not compress atomic metadata, we compressed the metadata-only with MMTF and then added that to the PNG sizes from PIC. All uses of MMTF were followed-up with standard Gzip compression on the MMTF file, as is standard, whereas BCIF is already a fully compressed file format. All sizes are in kilobytes (1000 \(\times\) bytes). We also include the RMSD for PIC coordinates here. Lastly, unfortunately, most of the proteins we chose for our benchmark were discontinuous, or had other quirks, so Foldcomp 0.0.5 was unable to compress them after running for 24 h. However, Foldcomp does substantially better on both 2jan and 4v60 than all but MMTF-reduced (C\(\alpha\)-lossy), which not only decreases precision but also only keeps the alpha carbons

ISSN: 1471-2105