research papers
Protein Data Bank depositions from synchrotron sources
aBiology Department, Brookhaven National Laboratory, USA
*Correspondence e-mail: jiang@bnl.gov
A survey and analysis of Protein Data Bank (PDB) depositions from international synchrotron radiation facilities, based on the latest released PDB entries, are reported. The results (https://asdp.bnl.gov/asda/Libraries/ ) show that worldwide, every year since 1999, more than 50% of the deposited X-ray structures have used synchrotron facilities, reaching 75% by 2003. In this web-based database, all PDB entries among individual synchrotron beamlines are archived, synchronized with the weekly PDB release. Statistics regarding the quality of experimental data and the refined model for all structures are presented, and these are analysed to reflect the impact of synchrotron sources. The results confirm the common impression that synchrotron sources extend the size of structures that can be solved with equivalent or better quality than home sources.
1. Introduction
The Protein Data Bank (https://www.rcsb.org/ ; Berman et al., 2000; Sussman et al., 1998) is the unique internationally recognized database for the depositing and archiving of biological macromolecular structures. By 31 December 2003, the PDB had released 23792 protein structures. Of these, 17102 entries were X-ray structures. Conventional (`home') X-ray sources (e.g. the rotating-anode-based sources) have been used for years to determine X-ray structures. Synchrotron radiation sources became available for structural biology in the 1980s, and have had a significant impact since the early 1990s. The availability of international synchrotron facilities was reviewed and summarized in Table 1 of Helliwell (1998). Because of the continual gradual construction of new beamlines for macromolecular crystallography, synchrotron sources are now widely available, and many depositions to the PDB depend on these sources. We have searched all beamlines that make depositions to the PDB, and developed an automatically updated database called PXLIB (https://asdp.bnl.gov/asda/Libraries/ ) to track what we believe is an objective measure of this productivity, which perhaps will be a useful tool for many in both the synchrotron and structural biological communities. We report on the work to produce the database, and also describe analyses on the data regarding the relative quality of structures determined at synchrotron X-ray sources.
The completion of the Human Genome Project at the end of the 20th century stimulated a new biological initiative known as `structural genomics' (SG). The goal of SG is systematically to determine the structures of proteins and important macromolecules, with the selection of the target proteins being guided somehow by the knowledge of the genomes of key organisms. This will yield a large number of `representative' protein structures in the near future. In North America the NIGMS/NIH-funded Protein Structure Initiative (https://www.nigms.nih.gov/psi/ ) was started in the year 2000 to develop nine SG centers (Table 1). International efforts were also established (https://www.isgo.org ); a SG issue was published in Nature Structural Biology in 2000 to summarize the worldwide program. The SG centers are focusing not only on producing protein structures but also on the development of techniques, software and apparatus to `pipeline' or automate the process of macromolecular (Chance et al., 2002; Holton & Alber, 2004; Brunzelle et al., 2003). Most SG centers are tied up with synchrotron facilities for the high-throughput structure determinations. Because it seems likely that the SG effort will become a particularly rich source of synchrotron-linked PDB entries, we pay particular attention to this advance. Table 1 lists the synchrotron sources involved with the US SG centers; https://www.rcsb.org/pdb/strucgen.html lists worldwide SG centers.
|
2. Methods in the development of PXLIB
The PDB began to collect information on X-ray experimental details in a comprehensive way in 1995. Only a few records can be found on synchrotron radiation information from the PDB database earlier than 1995, although the PDB was started in 1972 (Bernstein et al., 1977), and a substantial amount of data were collected at synchrotron sources during the period 1985–1995. Initially, the information about synchrotron radiation source and beamline was placed in records labelled `REMARK 18'. By 1995 the PDB was collecting and annotating X-ray experimental details more formally; this evolved into the `REMARK 200' record (Protein Data Bank Contents Guide, 1996). `REMARK 200 SYNCHROTRON' indicates that the X-ray data were produced by synchrotron radiation; `REMARK 200 RADIATION SOURCE' indicates whether the radiation source is from a home source (e.g. a rotating anode) or a synchrotron facility site (e.g. NSLS); `REMARK 200 BEAMLINE' indicates the particular beamline on which the data were collected. When two or more beamlines have been used, all beamline identifications should be presented in the deposition (separated by semicolons). In this study we parsed the latest released PDB entries and extracted all necessary information from the PDB flat files.
BIOSYNC (Structural Biology Synchrotron Users Organization, https://biosync.sdsc.edu )lists all beamline names in the USA but it does not provide beamline names for overseas. We have identified 82 synchrotron beamlines worldwide that have made depositions to the PDB, as listed in Table 2. Some beamline names have changed over time and non-standard beamline identifiers and multiple names have been deposited in the PDB entries. This leads to some difficulties in coding a program/script. We collect and include all possible alternative names and obsolete IDs in the brackets as `synonyms' to a `representative' beamline. Table 2 lists the `representative' beamlines for each synchrotron facility site. The detailed `synonyms' for each beamline are posted on the web site (https://asdp. bnl.gov/asda/Libraries/pdb_statis/ ). The `representative' beamline may not be a single physical beamline. For example, at APS, sector 22 has two physical beamlines, 22ID and 22BM. Since 22BM has not yielded any PDB deposition, `22ID' in Table 2 represents both beamlines 22ID and 22BM. Some beamlines were closed, for example, LURE at ORSAY in France. However, the depositions that were made in the past, and there may be more depositions in the coming years, will still be present in the table. If a PDB deposition is received from a new beamline (after confirmation with the synchrotron source), the new beamline will appear in the table on the web site.
|
In order to catch the latest entries released from the PDB, we have installed a mirrored PDB database at BNL. The mirrored PDB database is updated every week on Wednesday. The PXLIB script parses every PDB file to extract the data that are needed for the statistics. Table 3 lists the information that is extracted and collected in statistics files. We count every occurrence of beamline IDs in all PDB depositions as the `credits' for the synchrotron facilities. If multiple beamline IDs appear in the same record (separated by a semicolon), we count each occurrence as an equal fractional credit. For example, if both X25 (NSLS) and 19ID (APS) appeared on the same record of `REMARK 200 BEAMLINE', both X25 and 19ID will have half a credit. In order to identify the unique beamline ID, we first abbreviate the entry by removing or adding prefixes and insertions, such as `-', `BL', `PX' or `STATION', then we map it to the unique site ID and beamline ID as listed in Table 2. We make necessary corrections if some obvious errors are found in the PDB file. For example, `SSRF' should be `SSRL', `X12C = PF' should be `X12C = NSLS', `BL711 = PF' should be `I711 = MAXLAB'. There are some quoted IDs for beamlines but no indication of the synchrotron facility site. In this case, we found out the source ID from Table 2. Some PDB entries indicate the synchrotron facility site only but no information about the beamline; we denote them as in the `UNC' (unclaimed) beamline for this facility source site.
|
PXLIB runs automatically on every Wednesday in synchrony with the PDB release. Therefore, the tables in the `Latest Release' will automatically update every week. The data in the past weeks will be pushed down to `Archives'. Meanwhile, a conversion script generates HTML pages to update the PXLIB web site. PXLIB also builds up a gallery of PDB images of structures for each beamline. The program PXLIB is a set of scripts written in the PERL and C-SHELL languages. The source codes are available and can be downloaded from https://asdp.bnl.gov/asda/Libraries/ . PXLIB consists of two major steps: (i) extracting data from the mirrored PDB database, `run_pdb_statis1.csh'; (ii) performing analysis and statistics on the extracted data, `run_pdb_statis2.csh'.
3. Results and analysis
3.1. The numbers of PDB depositions from synchrotron sources
The numbers of PDB depositions from all synchrotron beamlines and various statistics derived from these have been posted at https://asdp.bnl.gov/asda/Libraries/pdb_statis/ . The numbers in Table 4 are an example of the sort of data that can be derived; one can see that the tables on the PXLIB web site are referenced to the year when the PDB deposition was made. There is a policy at the PDB that the depositor is allowed to demand that the entry be held for up to one year, waiting for publication of the structure, before being released. Since there are some depositions that had been made in the deposition year but not yet released, the numbers in the current deposition year will be partial counts. Table 4(a) summarizes the PDB depositions from synchrotron sources as of 31 December 2003. The column headed XRAY is the total number of X-ray structures in the deposition year (since 1995), that headed SYNC is the total PDB depositions from synchrotron sources, and that headed HOMES is the PDB depositions from home sources (including unknowns). In Table 4(b) the total SYNC numbers are broken down to the sources from the US (North America), EURO (Europe), ASIAN (Asia) and S.AMER (South America). One may also browse the PXLIB web site to find data for individual beamlines.
|
It is clear that by 1999 the synchrotron facilities contributed more than 50% of X-ray structures. The fraction leaps to 60% in 2000, and it continues a 5% growth annually. According to Table 4(a), PDB depositions from synchrotron sources will be more than 75% for year 2003. Since 1995, synchrotron sources from the US and Europe have contributed to 4505 and 3944 PDB structures, respectively (Table 4b). By the end of December 2003, the average depositions from the US (35 beamlines) and Europe (30 beamlines) were almost equal (130 depositions per beamline). Fig. 1(a) compares the number of depositions of each synchrotron facility site from US sources; Fig. 1(b) plots the depositions from European synchrotron sources, and Fig. 1(c) shows the depositions from Asian synchrotron sources. Since 1995, in the US the NSLS made the largest number of contributions to the PDB until 2002, at which point the third-generation APS took the lead. ESRF in Europe and SPring-8 in Asia are leading the PDB depositions.
There is a rich source of information available on the PXLIB web site (https://asdp.bnl.gov/asda/Libraries/pdb_statis/ ). Selecting `Latest Release' will present an overview of total PDB depositions from different areas; from there, selecting the location and the synchrotron facility IDs will lead to the tables for individual sites and beamlines. Then, within that, selecting a beamline will give a table of entries. Selecting `Gallery' allows one to break out depositions from each facility and to display thumbnail images of all entries from each beamline; selecting `Primary Citation' will list all publications that are related to this beamline; selecting the PDB ID code will lead to the PDB explore page (https://www.rcsb.org/pdb/cgi/explore.cgi?pdbid= 1abc ) which allows easy visualization of the contents of the PDB entry and the model using Protein Explorer (Martz, 2002). There are also alternate text forms rather than HTML, and raw data is provided in different formats. The source code is available from the web site, and potential users may contact the first author for assistance.
3.2. Lag time
The lag time is the difference between the date of data collection and the date of PDB deposition, in months. Table 5 shows, for each year, the average lag time and number of structures deposited for synchrotron and home sources. There is a clear difference between the lag times for structures from synchrotron sources and home sources: 18 months and 25 months, respectively. Evidently the structures that came from synchrotron facilities somehow seemed to be more urgent, or perhaps were easier to complete, and therefore were deposited sooner. The average lag time over the PDB depositions for each beamline is also calculated and is available from the PXLIB web site.
|
3.3. Statistics on resolution and the size of the molecular structure
Various quality statistics for X-ray structures from the latest released PDB entries are summarized in Tables 6 and 7. It is intriguing to note that in Table 6 the mean values of resolution limit and both R values are essentially identical for synchrotron and home sources. The difference is that the typical structure solved at a synchrotron source is more than half as large again as that solved on a home source, i.e. 738 residues and 453 residues, respectively. Several beamlines shown an average size of more than 1000 residues with a large standard deviation, which indicates that the beamline might have recorded data for larger macromolecular complex structures, such as ribosome. We analyse this difference later, but because of this similarity we combine the two for several analyses. The distribution of the reported resolutions over all deposited X-ray structures is plotted in Fig. 2. The average resolution is 2.2 Å with a standard deviation of 0.5 Å over 17144 structures. More than 90% of structures have the data falling in the resolution range between 1.5 Å and 3.0 Å. Only 6% of structures have a higher resolution than 1.5 Å, and 4% of structures have resolutions lower than 3.0 Å. The mean and standard deviation for each synchrotron site and beamline look similar except for a number of beamlines that have a few large structures (with 7.0 Å to 8.0 Å resolution); a corollary of this is that several beamlines have a larger average size than the others.
|
|
3.4. Statistics on R values and the free R values
The crystallographic R value and free R value (Brünger, 1992, 1997) are reported in each PDB deposition to define how well coordinates of the model fit the X-ray data. Fig. 3 displays the distribution of the reported R values and free R values over 14746 structures. Each spot represents a structure. A number of free R values were reported less than R values in the plot (below the diagonal line y = x). These free R values might not be correct owing to the inefficient `cross-validation'. Free R values and R values are highly correlated. The over all reported X-ray structures is 0.814 (see Table 8). The R value has a mean of 0.20 and a standard deviation of 0.03 and the free R value has a mean of 0.25 and a standard deviation of 0.04 (see Table 7). The means of R values and free R values as a function of the reported resolution are plotted in Fig. 4. Error bars represent the associated standard deviations. The resolution bin is set to 0.2 Å. The number of the R/free-R values is significant in the resolution bins from 0.8 Å to 3.6 Å (dmin). The standard deviations are smooth in the range between 0.027 and 0.037 for both the R values and free R values. The dependency of both R value and free R value on the resolution is clearly shown. We have further analysed the relation between the free R value and the resolution based on these statistics, which will be reported in a separate paper (Jiang, 2004).
|
3.5. Statistics on X-ray data qualities and coordinate model qualities
Some statistics representing the quality of the raw data and of the refined models, averaged over all structures, are summarized in Table 7. The completeness of the data measured has a mean of 93% with a standard deviation of 8%, and does not depend on resolution. The redundancy has a mean of 4.9 with a standard deviation of 3.6 (Fig. 5) and the distribution does not depend on resolution. 〈I/σ(I)〉 has a mean of 16 with a standard deviation of 10 (Fig. 6) and its distribution also does not depend on resolution. The statistical properties of the reported R-merge (12480 cases) and the reported R-sym (only 5487 cases) are very similar. R-merge has a mean of 0.074 with a standard deviation of 0.05 (Fig. 7a) while R-sym has a mean of 0.077 with a standard deviation of 0.06. The distribution of R-merge has a small dependence on resolution. The means of R-merge with associated standard deviations are plotted as a function of resolution in Fig. 7(b).
The r.m.s. bond-length deviation from the ideal has a mean of 0.012 Å with a standard deviation of 0.008 Å (Fig. 8a), and the r.m.s. bond angle deviation from the ideal has a mean of 1.7° with a standard deviation of 0.8° (Fig. 9a). In Fig. 8(b) and Fig. 9(b) are plotted the means and standard deviations against the resolution; neither depends on the resolution. The bond-length r.m.s.d. has a higher mean in the high-resolution structures (dmin < 1.5 Å) and a smooth lower mean (around 0.01) through the lower-resolution range. From this it can be interpreted that the lower-resolution (dmin > 1.5 Å) structures were heavily restrained to achieve a good geometry of the model (Engh & Huber, 1991). Fig. 10(a) shows the distributions of the overall B-factors. The average of the overall B-factors on 11736 structures is 31 Å2 with a standard deviation of 16 Å2. The means and standard deviations of the B-factors as a function of resolution are given in Fig. 10(b). It is no surprise that as dmin increases (lower resolution) the mean of the B-factors increases; the standard deviation also increases. The coordinate errors estimated by the Luzzati plot (Luzzati, 1952) derived from the cross-validated free R value have a mean of 0.33 Å with a standard deviation of 0.11 Å over 5447 structures; the coordinate errors estimated (cross-validated) by the σA plot (Read, 1986, 1990) have a mean of 0.33 Å with a standard deviation of 0.18 Å over 5004 structures; the estimated coordinate errors by ESU [estimated based on the free R value given by REFMAC (Murshudov et al., 1997)] have a mean of 0.18 Å with a standard deviation of 0.12 Å over 1081 structures.
We have computed the correlation coefficients (cc) between pairs of some statistical properties over all released X-ray structures from the PDB (Table 8). Most data and model qualities are correlated to the resolution except for r.m.s. bond length and angle deviations from ideal. The estimated coordinate errors are highly correlated with the resolution, free R value and B-factor (cc > 0.5), but only slightly correlated with R-merge (cc > 0.21).
3.6. Differences between synchrotron and home sources
Tables 6, 7 and 8 show any differences that might arise between data taken at synchrotrons and with home sources. As already mentioned, the data in Table 6 reveal the slightly surprising statistic that the most dramatic difference between the SYNC- and HOMES-based structures is the size of the structure, not the final quality. This is borne out in the thorough analysis of data and model quality in Table 7, where the only significant differences are shown in bold type. These suggest that the average number of reflections from synchrotron sources is much higher than those from home source, 57928 versus 27770. Also, since the average resolution limit between SYNC and HOMES is the same (Table 6), the average number of refined atoms follows: 6071 versus 3731. However, this relationship is not strictly true since, as the SYNC data tend to be slightly more complete, the ratio of atoms over reflections is slightly smaller in SYNC than in HOMES. Since it is this parameter-to-data ratio that defines the quality of a final refined structure, one might expect the SYNC structures to be slightly more accurate at a given resolution.
The other interesting discrepancy in Table 7 is that the average B-factors are significantly larger for SYNC data than for HOMES data. Not only are SYNC-derived structures larger on average, but they are also harder to determine because the B factors are larger. This set of statistics [i.e. equivalent resolution and R value, but bigger structures and weaker high-resolution data (larger numbers of reflections)] represents the crux of the power of synchrotron radiation for macromolecular crystallography.
It is worth noting that the correlations (Table 8) between R-value/R-free and the reported resolution are much higher for SYNC than for HOMES (0.63 verses 0.34 on R value, 0.70 verses 0.51 on R-free). B-factors tend to be more highly correlated with SYNC-derived resolution and R values than with HOMES-derived ones (Table 8).
3.7. X-ray crystallographic methods
X-ray crystallography methods are also reported in the PDB entry. The information can be extracted from the record of `REMARK 200 METHODS USED'. However, the record is a text string that is difficult to parse, and about a quarter of entries do not have such information. We abbreviate the commonly used methods (https://asdp.bnl.gov/asda/Libraries/pdb_statis/latest/xme/METH.html ) and then make necessary corrections and interpretations. We divide the methods into four major categories: experimental phasing methods, and miscellaneous. The percentages of these four categories are 59%, 23%, 1% and 17%, respectively. The experimental phasing methods can be divided into two groups: the single methods (83%) and the combinations (17%). The single methods are composed of four individual phasing methods, multiple (MIR, 31%), single (SIR, 10%), multiwavelength anomalous diffraction (MAD, 52%) and single-wavelength anomalous diffraction (SAD, 7%). The combinations are the combination of and Table 9 summarizes all reported X-ray crystallographic methods. Fig. 11 shows the numbers of reported experimental phasing methods (individual) referenced to the deposition year. One can see that the number of MAD structures has increased dramatically (Ogata, 1998), and it doubled in one year, from 2001 to 2002. One might also note that the ratio of SAD to MAD structures seems to be increasing precipitously. This single-wavelength anomalous method works best with tunable synchrotron radiation and the other qualities of synchrotron facilities: well collimated beams to provide good signal-to-noise, and state-of-the art detection systems.
|
4. Discussion
4.1. Number of PDB depositions from synchrotron sources
In this survey, all counts are based on the PDB depositions beginning in 1995. A substantial number of PDB depositions earlier than 1995 are excluded because the PDB did not systematically collect information about the X-ray source. Some depositors might have misquoted the beamline IDs, and some errors exist in the primary PDB database. There may be a couple of beamlines that are in operation but do not appear in these statistics because they have not yet made a PDB deposition. We will add these or new beamlines once the PDB deposition is made.
Non-standard beamline IDs and the evolution of beamline names make it difficult to extract information from the primary PDB database. We have been communicating with the scientists at synchrotron facility sites to verify the deposited information on beamlines and we have made necessary corrections if there are errors. PXLIB will provide more accurate information on synchrotrons than the primary PDB database. We suggest that the synchrotron source community should have a standardization in defining beamline names. The users of synchrotron sources should be reminded to provide accurate beamline information to the PDB when they make a deposition.
[However, the validation of synchrotron beamline information has involved checks and cross-checks with the synchrotron radiation facilities themselves made by ourselves and include the further suggestions of the referees to whom we are grateful. If any errors remain we welcome corrections that will, if necessary, lead to an updated analysis that will be submitted for publication to the Journal of Synchrotron Radiation at a later date. As part of this process, input from facilities on these matters should be addressed to the corresponding author, jiang@bnl.gov.]
4.2. Impact on the PDB of SG and high-throughput methods
The SG approach differs from traditional structural biology by its organized `pipeline'. The organized `pipeline' includes all procedures from protein target selection, cloning and expression, crystallization, data collection, ). Several synchrotron beamlines have been supported by SG and reconstructed with a crystal-mounting mechanism (`robot') for high throughput. According to the report from PSI (Annual PSI Meeting 2003, NIGMS, Bethesda, MD), the nine PSI SG centers have determined 537 structures and deposited 335 structures to the PDB since funding was started in 2000 (Table 1, personal communications). Although we did not observe a big jump in the total numbers of PDB depositions since the start of SG in 2000, several beamlines participating in SG projects have seen a burst in PDB depositions. For example, 19ID at APS (MWSG) had more than 100 depositions in 2002 and 2003. Not only will the SG approach produce more structures, but also new technologies will speed up the whole process of For example, the lag time on beamline 31ID/APS (SGX) is only three months and on X06SA at SLS is only six months. The average lag time (not shown) for selective beamlines that involve SG centers (as listed in Table 1) is less than 12 months compared with 18 months over all beamlines. More importantly, the SG approach produces more `unique' structures (less than 30% sequence identity to the others) and more `new' protein folds, because the new approach scans the whole genomes of all kinds in order to explore `new' structures. Experimental phasing methods, particularly selenomethionine MAD/SAD phasing methods, are the major methods for SG because homologue models are not available for molecular replacement.
and function research. The goal of SG is not only to determine the structures of proteins systematically, but also to develop high-throughput and automation methods that can be used in a production line. All of the SG centers are closely related to at least one synchrotron facility (Table 1Acknowledgements
The author is grateful to Lonny Berman for contributions to this paper. The authors acknowledge support from NCRR/NIH grant P41-RR12408 (PI: R. M. Sweet) and NIGMS/NIH grant P50-GM62529 (PI: S. K. Burley). We would like to thank Lonny Berman, Mike Becker and Mark Chance for the consensus on the `equal fractional credit' in counting multiple beamlines. Thanks also to Nancy Manning for the clarification on the `REMARK' records, and Zheng Lin for help on the mirrored PDB database.
References
Berman, H. M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T. N., Weissig, N., Shindyalov, I. N. & Bourne, P. E. (2000). Nucl. Acids Res. 28, 235–242. Web of Science CrossRef PubMed CAS
Bernstein, F. C., Koetzle, T. F., Williams, G. J. B., Meyer, E. F. Jr, Brice, M. D., Rodgers, J. R., Kennard, O., Shimanouchi, T. & Tasumi, M. (1977). J. Mol. Biol. 112, 535–542. CrossRef CAS PubMed Web of Science
Brünger, A. T. (1992). Nature (London), 355, 472–475. PubMed Web of Science
Brünger, A. T. (1997). Methods Enzymol. 277, 366–396. CrossRef PubMed CAS Web of Science
Brunzelle, J. S., Shafaee, P., Yang, X., Weigand, S., Ren, Z. & Anderson, W. F. (2003). Acta Cryst. D59, 1138–1144. Web of Science CrossRef CAS IUCr Journals
Chance, M. R., Bresnick, A. R., Burley, S. K., Jiang, J. S., Lima, C. D., Sali, A., Almo, S. C., Bonanno, J. B., Buglino, J. A., Boulton, S., Chen, H., Eswar, N., He, G., Huang, R., Ilyin, V., McMahan, L., Pieper, U., Ray, S., Vidal, M. & Wang, L. K. (2002). Protein Sci. 11, 723–738. Web of Science CrossRef PubMed CAS
Engh, R. A. & Huber, R. (1991). Acta Cryst. A47, 392–400. CrossRef CAS Web of Science IUCr Journals
Helliwell, J. R. (1998). Nature Struct. Biol. Suppl. 5, 614–617. Web of Science CrossRef CAS
Holton, J. & Alber, T. (2004). Proc. Natl Acad. Sci. USA, 101, 1537–1542. Web of Science CrossRef PubMed CAS
Jiang, J.-S. (2004). Acta Cryst. D60. To be submitted.
Luzzati, V. (1952). Acta Cryst. 5, 802–810. CrossRef IUCr Journals Web of Science
Martz, E. (2002). Trends Biochem. Sci. 27, 107–109. (https://www.protein explorer.org/ .)
Murshudov, G. N., Vagin, A. A. & Dodson, E. J. (1997). Acta Cryst. D53, 240–255. CrossRef CAS Web of Science IUCr Journals
Ogata, C. M. (1998). Nature Struct. Biol. Suppl. 5, 638–640. Web of Science CrossRef CAS
Protein Data Bank Contents Guide (1996). Protein Data Bank Contents Guide: PDB Format Description, Version 2.1, 25 October 1996, https://www.rcsb.org/pdb/docs/format/pdbguide2.2/guide2.2_frame.html .
Read, R. J. (1986). Acta Cryst. A42, 140–149. CrossRef CAS Web of Science IUCr Journals
Read, R. J. (1990). Acta Cryst. A46, 900–912. CrossRef CAS Web of Science IUCr Journals
Sussman, J. L., Lin, D., Jiang, J.-S., Manning, N. O., Prilusky, J., Ritter, O. & Abola, E. E. (1998). Acta Cryst. D54, 1078–1084. Web of Science CrossRef CAS IUCr Journals
© International Union of Crystallography. Prior permission is not required to reproduce short quotations, tables and figures from this article, provided the original authors and source are cited. For more information, click here.