protein topology

Andrew C.R. Martin, The ups and downs of protein topology; speedy comparability of protein construction, Protein Engineering, Design and Choice, Quantity 13, Subject 12, December 2000, Pages 829–837, https://doi.org/10.1093/protein/13.12.829

 

Summary

 

Introduction

As quickly growing portions of recent construction information develop into accessible, the classification of protein folds is changing into an increasing number of necessary. When a brand new construction is solved, one needs to ask the query of whether or not this fold has been seen earlier than. If the fold is likely one of the generally occurring superfolds (e.g. immunoglobulin fold, TIM barrel, αβ-plait, Rossmann fold), then it could possibly typically be acknowledged by eye, however the much less frequent folds are tougher to acknowledge. Outdoors these superfolds, there may be typically a one-to-one mapping between fold and homologous household. In some circumstances, distant homologues might solely be recognised within the gentle of structural similarity permitting inference of purposeful data. Automated servers to reply the classification query have just lately been developed and depend on construction comparability packages.

Sadly, detailed computerized construction comparability is a time-consuming course of. The double-dynamic programming algorithm utilized in SSAP (Taylor and Orengo, 1989) on the atomic coordinate degree is especially laptop intensive making it impractical to run a full scan of a protein of any measurement in opposition to a library of recognized folds. An experimental `CATH-server’ at College Faculty London (Orengo et al., 1998) has made use of sequence screening and a hierarchical enlargement scheme via the consultant ranges within the CATH area classification database (Orengo et al., 1997) to scale back the search, however within the worst-case situation, it may nonetheless be essential to scan in opposition to greater than 3000 near-identical sequence representatives. It’s estimated that ~2 weeks of laptop time could be required to scan a big protein area reminiscent of a TIM barrel on this manner and that is clearly impractical to be used as a server. The preliminary purpose of this work due to this fact was to develop a speedy approximate technique for protein fold comparability which could possibly be used as a pre-screen for SSAP.

The CATH classification of protein area buildings is a hierarchical classification encoding class (C), structure (A), topology (T), homology (H), sequence household (S), near- similar sequence (N) and similar sequence (I). Using the phrase `topology’ on this description is one thing of a misnomer. After we consult with the topology of a protein, what we typically imply is the three-dimensional fold. Extra strictly, for a given spatial association of secondary construction parts (SSEs), the topology describes how these parts are related.

The dictionary definition of `topology’ is `the elements which stay unchanged as an object undergoes a steady deformation’. By way of protein secondary construction, the true topology is solely the sequence of SSEs, i.e. if one imagines with the ability to maintain the N- and C-terminal ends of a protein chain and pull it out straight, the topology doesn’t change regardless of the protein fold (offering no knots are shaped within the folded protein). Right here, we describe this because the `main topology’ whereas, by analogy with main and tertiary construction, the protein fold is described because the `tertiary topology’. A main topology string is a sequence of E and H characters representing β-strand and α-helix in DSSP notation (Kabsch and Sander, 1983).

When a sequence for a protein of unknown construction doesn’t present apparent sequence homology with a protein of recognized construction, fold recognition is usually used to offer clues as to the three-dimensional construction. There are two frequent approaches to this downside: one is `threading’ which assesses how properly a sequence is accommodated inside a three-dimensional construction (e.g. Jones et al., 1992); the opposite is alignment of a predicted sequence of secondary construction assignments (Rost et al., 1997) or SSEs (Russell et al., 1996) in opposition to a fold library encoded on this type. The latter fold recognition technique is due to this fact making the idea that the tertiary topology may be predicted from the first topology. An evaluation of the prevalence of given main topologies in numerous tertiary topologies is introduced.

If this assumption is true, then it must also be attainable to make use of main topology strings for three-dimensional construction comparability. Nonetheless, given the additional data accessible inside a three-dimensional construction, one can introduce an intermediate degree of topology `secondary topology’ which comprises extra data (reminiscent of SSE path, proximity, accessibility and size of the weather and the loops which join them) to enhance the mapping between these decrease ranges of topological description and tertiary topology (i.e. protein fold).

Plenty of different strategies of protein construction comparability have made use of a simplified illustration of protein construction for the reason that idea was first recommended by Sheridan et al. (1985). For instance, Mitchell et al. (1990) appeared for frequent sub-structures by utilizing a linear illustration of helices and strands to create a graph amenable to comparability utilizing subgraph isomorphism. Kleywegt and Jones (1997) described a protein as a set of SSEs along with variety of residues and bodily size of every SSE and used matrices of distances between the centres of the SSEs and cosines of angles between the path vectors of the weather. These representations have been then in comparison with discover related motifs.

Different strategies related in precept to TOPSCAN embody FAST-SSAP (Taylor and Orengo, 1989), a constraint programming (CP) technique primarily based on TOPS diagrams (Gilbert et al., 1999) and TOP, a least-squares becoming technique (Lu, 2000). FAST-SSAP makes use of double dynamic programming to align vectors representing the SSEs to check two buildings. The CP technique makes use of a four-letter alphabet to symbolize the 2 kinds of secondary construction going both up or down and provides details about chirality and hydrogen bonds between SSEs. Comparability is carried out by discovering a most frequent template and scoring the 2 buildings in opposition to that template. TOP represents every SSE by its two end-points after which performs iterative least-squares becoming of this small variety of factors to seek out an optimum match.

The foremost distinction between the tactic introduced right here and different strategies is that the illustration of SSEs is only symbolic. This has a definite velocity benefit and implies that solely a single algorithm (dynamic programming) is required for performing the comparability of buildings. No comparisons are wanted on the degree of coordinates or distances. Right here an evaluation of protein fold similarity on the main and secondary topology ranges is introduced and the flexibility of those to foretell similarity on the tertiary topology (protein fold) degree is assessed. Implications of the outcomes for fold recognition and secondary construction prediction are mentioned.

 

Supplies and strategies

TOPSCAN reduces a protein to a topology string which may symbolize the construction as a string of letters encoding merely the first topology (a two-letter alphabet) or the secondary topology utilizing numerous extra information from the 3D construction. For instance, by encoding path data with the secondary construction, a 12-letter alphabet is used.

A easy Needleman and Wunsch dynamic programming algorithm (Needleman and Wunsch, 1970) is then utilized to check two such strings and a share similarity rating is calculated. The proportion similarity rating is calculated as the proportion of the upper rating achieved by scoring every of the 2 sequences in opposition to itself. Alternatively, somewhat than evaluating two buildings, a library of topology strings could also be pre-built from the Protein Databank (or a consultant subset) and a construction might then be scanned in opposition to this library.

The TOPSCAN program itself is applied in C. All evaluation and management of the packages was achieved utilizing units of scripts written in Perl and pushed from a relational database of the CATH information applied utilizing the freely accessible object-relational database PostgreSQL (http://www.postgresql.org). Evaluation was carried out on a twin PentiumIII-450 machine working RedHat Linux 6.0.

Datasets

All evaluation was carried out utilizing datasets derived from the CATH classification of protein area folds (Orengo et al., 1997). In CATH, consultant buildings are assigned at T, H, S and N ranges. For evaluation of parameters for use in this system, the representatives from the H degree (Hreps) from CATH v1.6 have been chosen, however from this set, any buildings now obsoleted from the protein databank (Berman et al., 2000) and any NMR buildings have been eliminated. This gave a set of 914 domains representing a set of non-homologous protein domains.

For testing the efficiency of this system, an identical set was created utilizing the CATH Nreps resulting in a set of 3124 domains. Every Nrep is a consultant of a close to similar group of sequences having ≥95% sequence identification. These have been used to construct the libraries in opposition to which every of these 3124 domains was examined (matches in opposition to self have been all the time ignored in calculating outcomes).

Creating main topology strings

TOPSCAN permits secondary construction to be calculated from a three-dimensional construction utilizing both DSSP (Kabsch and Sander, 1983) or STRIDE (Frishman and Argos, 1995). For the analyses introduced on this paper, STRIDE was used. From these assignments, areas of β-sheet (Kabsch and Sander project, E) and of α-helix (Kabsch and Sander project, H) are extracted. Solely steady areas of no less than a specified variety of residues with the identical project are chosen. This produces the first topology of the protein equal to a string of E and H characters the place one character represents one full strand or helix. This system additionally permits one to deal with 310-helix assignments as α-helix as is steadily finished in secondary construction prediction.

Creating secondary topology strings

To extend the knowledge content material of the topology string, numerous data from the three-dimensional construction may be included right into a secondary topology string.

For simplicity and readability of explanations, topology strings are described right here as vectors of characters. In apply, integer vectors are used to permit greater than 52 descriptors. This additionally supplies a technical simplification as lookups within the scoring matrix could also be made just by the integer offsets for the 2 descriptors being in contrast.

Secondary construction aspect (SSE) path

The top-points of every SSE within the main topology are present in three dimensions and the vector between them is calculated. The path of the vector is grouped into one among six lessons relying on the biggest part of the vector (i.e. constructive or adverse x, y or z). That is equal to saying the aspect factors up, down, left, proper, ahead or again. The encoding is summarized in Desk I.

The scoring matrix used for the dynamic programming algorithm is predicated on the scoring scheme proven in Desk II. Though considerably arbitrary, it seems to work properly: the identical secondary construction in the identical orientation scores highest, completely different orientations rating worse and completely different secondary construction sorts obtain a lot decrease scores.

The beginning premise was that for a similar kind of secondary construction, one needs to assign three scores: roughly the identical path (0° ≤ Δ ≤ 45°) scores 10, roughly at right-angles (45° ≤ Δ ≤ 135°) scores 5, roughly reverse (135° ≤ Δ ≤ 180°) scores 2. Nonetheless, due to the boolean definition of whether or not a vector is in a given quadrant, it’s attainable that vectors really level in very related instructions though they’re in numerous quadrants (for instance, one factors at +44° whereas one other factors at +46°; as a result of the definition of quadrant depends upon the biggest part of the path vector, the quadrant boundaries are at ±45° and ±135°). Selecting pairs of random numbers between 0–90 and 90–180 and plotting the distribution of the variations exhibits an upside-down V-shaped curve centred round 90 (information not proven). Two vectors in adjoining quadrants will really be inside 90° of each other 50% of the time. The precise rating used for adjoining quadrants is due to this fact the median of 10 and 5 (7.5 rounded as much as 8).

When two vectors are assigned to reverse quadrants, they’ll by no means be nearer than 90°. Selecting pairs of random numbers between 0–90 and 180–270 provides an similar distribution centred round 180. The angle between two vectors can by no means be higher than 180°, so each angle <180° noticed between vectors in reverse quadrants may also be seen in vectors assigned to adjoining quadrants. In reverse quadrants, the angle between two vectors is <135° (our cutoff for saying two vectors are roughly at proper angles) 12.5% of the time. The rating is due to this fact assigned as 2 + [(5 – 2)×12.5/100] = 2.375, which is rounded again all the way down to 2. For various secondary construction sorts, we make the straightforward arbitrary assignments of three, 1 and 0. A spot insertion penalty of 8 is used with no hole extension penalty. As a result of any pair of proteins is in an arbitrary relative orientation, the definition of `up' in a single protein might not correspond to `up' within the second. Due to this fact, one of many strings is permuted 23 instances, such that the dynamic programming algorithm comparability is carried out a complete of 24 instances (equal to the six sides of a dice, every of which can be 4 methods up). Desk III exhibits the modifications made to the encoding to attain rotations in regards to the x, y and z axes. Secondary construction aspect (SSE) proximity So as to add details about the packing of SSEs, the proximity of an SSE to the previous aspect is encoded. That is carried out as follows (see Determine 1). Given a component with endpoints C,D and a previous aspect with endpoints A,B, the minimal distances between factors A and B and the road described by CD are calculated. That is repeated with factors C and D to line AB. As every minimal point-to-line distance is calculated, the place on the road closest to the purpose is calculated (proven in Determine 1 with a black circle). If this level is between the 2 endpoints of the road and the gap is <12.0 Å, then we are saying that the aspect C,D is proximal to the previous aspect and encode this inside the topology strings. The N-terminal aspect doesn't have a proximity parameter assigned to it. The scores within the similarity matrix are multiplied by 0.75 and rounded all the way down to the following integer for mismatches in proximity. This additionally applies to the opposite gadgets encoded in secondary topology strings. Notice that every one cutoffs use single values, so the penalty of a mismatch (a discount of the rating by 25%) is just small.

Accessibility Utilizing the Hrep fold library, the imply residue accessibility was calculated for all residues in helices and in strands as assigned by STRIDE. This makes use of the algorithm of Eisenhaber and colleagues (Eisenhaber and Argos, 1993; Eisenhaber et al., 1995). Since absolute, somewhat than relative, accessibility is used, a component of amino acid composition can also be taken into consideration. The imply was calculated for every SSE after which averaged over all parts. The values obtained depend upon the minimal allowed size for an SSE: three residue helices, 52.8 Å2; 4 residue helices, 52.8 Å2; three residue strands, 33.8 Å2; 4 residue strands, 32.4 Å2. The values for helices additionally change if 310-helices are handled as α-helices (three residue cutoff, 56.5 Å2; 4 residue cutoff, 53.3 Å2). SSEs with imply residue accessibilities lower than these imply cutoffs are assigned as buried, others are uncovered. Factor size Imply helix and strand lengths have been calculated on the identical foundation utilizing STRIDE secondary construction assignments. The imply helix size is 12.5 residues whereas the imply strand size is 5.4 residues. If a component is longer than the suitable cutoff, it's outlined as `lengthy', in any other case `brief'. If 310-helix assignments are handled as α-helix, then the imply for helix size falls to 10.5 residues on account of the massive variety of brief 310-helices. The imply size of a 310-helix was additionally calculated and located to be 3.3 residues. Loop size Imply loop lengths have been calculated in the identical manner and assigned to the aspect which follows the loop. The primary SSE due to this fact doesn't have a loop size parameter assigned to it. The imply loop size was 6.9 residues. On this case, 310-helix assignments are handled as loop. If, as a substitute, they're handled as helix, the imply loop size was 5.7 residues.  

Outcomes and dialogue

Task of secondary construction (main topology)

The project of secondary construction is the important first stage. We have now discovered that the DSSP algorithm (Kabsch and Sander, 1983) may be over-sensitive to errors within the construction. For instance, NMR buildings usually present little secondary construction in DSSP assignments whereas an intuitive visible inspection exhibits substantial secondary construction. The DSSP algorithm works by assigning a cutoff to hydrogen-bonding energies and discovering patterns of hydrogen bonds attribute of secondary buildings. Any hydrogen bonds which fail to satisfy these standards are excluded and secondary construction project might due to this fact be lower than optimum.

The STRIDE software program from Frishman and Argos (1995) was designed to scale back this sensitivity. Nonetheless, utilizing STRIDE rather than DSSP makes little distinction to the general outcomes (information not proven). The TOPSCAN software program permits both DSSP or STRIDE secondary construction assignments for use and STRIDE was chosen for all additional evaluation.

We explored the results of setting the minimal required variety of consecutive residues assigned to a given secondary construction to a few and 4 residues. We additionally checked out treating 310-helix residues as the identical as α-helix (information not proven). This made little distinction when the size cutoff was set to 4, however made outcomes considerably worse when the size cutoff was set to a few. It is because most cases of true 310-helix are lower than 4 residues in size (as proven above, the imply size of a 310-helix is 3.3 residues). These are then handled as α-helix and in contrast with areas of true α-helix. This implies that treating 310-helix as α-helix just isn’t a great technique and this has implications for many who do that in secondary construction prediction (e.g. King et al., 2000). It seems that it might be higher to deal with 310-helix as coil in secondary construction prediction, notably when that is getting used for fold recognition.

Task of secondary topology

Along with secondary construction and the path of SSEs, the next elements have been thought of in defining secondary topology: accessibility, proximity, aspect size and loop size. The efficiency is assessed by way of p.c protection versus share error, particularly taking error charges of 1 and 5%. The dataset used was the CATH Nreps with NMR buildings excluded. Error and protection at a given TOPSCAN rating have been calculated as (variety of false hits/complete variety of hits) and (variety of true hits/complete true hits), respectively, and protection values at 1 and 5% error charges have been calculated by linear interpolation of the protection versus error plots within the ranges 0.5–1.75 and three.0–8.0, respectively. This offers a pattern of roughly eight factors evenly unfold above and beneath the interpolation level. These outcomes are summarized in Desk IV.

As may be seen in Desk IV, the very best protection outcomes are achieved with the secondary construction cutoff set to a few and with accessibility, proximity, aspect size and loop size data within the secondary topology strings. Curiously, on the 5% error price, loop size has a detrimental impact, presumably as a result of the loops are essentially the most variable issue between proteins adopting the identical fold.

Does main topology predict protein fold?

Utilizing the CATH classification, every protein fold containing a couple of near-identical sequence household (i.e. every CAT degree containing a couple of Nrep) was examined. Inside every of those 292 teams, the imply TOPSCAN rating was calculated for all pairs of Nreps utilizing solely the first topology (secondary construction cutoff size set to a few). The frequency of scores obtained is proven in Determine 2a. Unsurprisingly, the very best peak is for a rating of 95–100% (see Desk V), i.e. inside a protein fold, the first topology is mostly similar. As proven within the desk, related outcomes have been obtained when trivial matches between very small buildings containing fewer than 4 SSEs have been ignored.

Each Nrep inside every protein fold (CAT) was then in contrast with each Nrep of each different protein fold (a complete of 4 688 807 comparisons) and the distribution of the scores was plotted (Determine 2b). Initially one would count on that completely different folds can have a low rating on this comparability, however 14.2% of the comparisons rating >65% with 0.66% scoring >95% (see Desk V), the distribution peaking at a rating of 35–40%. Nonetheless, when one considers, for instance, 4-helix bundles as proven in Determine 3, it turns into apparent that there are a number of folds containing the identical set of SSEs. The determine exhibits two 4-helix bundles from bovine acyl-coenzyme A binding protein (Kragelund et al., 1993; PDB code, 1aca) and cytochrome c′ from Rhodospirillum molischianum (Finzel et al., 1985; PDB code, 2ccy). On the main topology degree, each could be described as HHHH (word that 2ccy has a brief phase of 310-helix between the second and third fundamental helices).

Thus, by figuring out solely the first topology (maybe from secondary construction prediction), there’s a excessive likelihood of matching the inaccurate protein fold with an similar main topology. These outcomes distinction with the protein taxonomy primarily based on secondary construction derived by Przytycka et al. (1999) which is predicated purely on main topology strings. They calculate a similarity rating on the premise of the frequent variety of residues matched when a pair of SSEs is aligned. From the outcomes introduced right here, it seems that the authors have been both fortunate to keep away from false constructive matches within the information set they used or that the scoring on the premise of size is an excellent filter.

Does secondary topology predict protein fold?

As with comparability of main topology, every protein fold containing a couple of near-identical sequence household was examined. Inside every of those teams, the TOPSCAN rating was calculated utilizing the secondary topology with a secondary construction size cutoff of three and together with accessibility, proximity, aspect size and loop size data (Determine 4a, Desk V). The distribution strikes to the left in contrast with utilizing main topology and the height for 95–100% is misplaced. That is largely a results of the one cutoffs used to outline secondary topology. Scores are artificially lowered the place two examples lie near, however on both aspect of, the cutoff. The distribution peaks are round 55–60%.

Comparability between protein folds (Determine 4b) now peaks at a rating of 15–20% and all of the high-scoring peaks are eliminated. Solely 0.15% of comparisons rating >65% (Desk V).

Referring once more to Determine 3, though the first topology for the 2 folds is similar (HHHH), by following the chain hint, it’s clear that the instructions of the helices are completely different (1aca, up, down, down, up; 2ccy, up, down, up, down) and incorporating this data into the secondary topology strings will distinguish between the 2 folds.

Curiously each the comparisons of main and secondary topology inside a fold present one comparability with a really low rating (<10%). These structures (domain 3 from PDB files 1dar (al-Karadaghi et al., 1996) and 1elo (Aevarsson et al., 1994)) were examined using RasMol and are shown in Figure 5. They represent one domain from elongation factor G (EF-G) from Thermus thermophilus strain HB8 complexed and uncomplexed with GDP, respectively. Whereas 1dar contains only β-strands, 1elo has two poorly defined strands and a helix. The sequences of these two are identical, yet the fold is different as a result of conformational flexibility of EF-G on binding nucleotide. It should also be noted that the structures are poorly defined in this region with some residues not visible in the crystal structures. Thus some of the differences may reflect errors in the structures although, since both structures are solved by the same group, it is unlikely that they would have built one with an α-helix and the other without unless there were a real difference. TIM barrels and Rossmann folds At the primary topology level, the TIM barrel and the Rossmann fold are somewhat similar—in both cases, they consist in the main of alternating β-strands and α-helices which form a core of β-sheet with helices on the outside. In the case of the TIM barrel the β-sheet is curved into a barrel whereas the Rossmann fold has a relatively flat sheet. Because the differences between the two occur in a plane perpendicular to the direction of the SSEs, the directional element of the secondary topology assignments is also similar. However, differences do occur in the accessibilities and lengths of the elements. It is interesting that in Rose's taxonomy based on secondary structure (Przytycka et al., 1999), the flavodoxin and flavodoxin-like Rossmann folds are well separated from the TIM barrels. However, one of the structures which they place in the middle of the TIM barrel cluster (2cmd: lactate and malate dehydrogenase) is split in the CATH classification into two domains, the first of which is assigned as a Rossmann fold. Table VI shows three examples of TIM barrels and Rossmann folds having almost identical primary topology. Although in these cases the secondary topology similarities are much lower, it can be seen that the scores are relatively high for such different protein folds although, in practice, they are comparable with the SSAP scores. Comparison of TOPSCAN with other rapid methods As cited in the Introduction, a number of other methods have used a reduced representation of protein structure to allow rapid comparison. The main difference between TOPSCAN and other methods is that the representation is purely symbolic and this simple approach allows a very significant speed advantage. It is approximately 25 times faster than the current implementations of FAST-SSAP (Taylor and Orengo, 1989) and CP (Gilbert et al., 1999) and is estimated to be approximately 30 000 times faster than normal SSAP. FAST-SSAP could be speeded by calculating the secondary structure vectors simply from the end-points of the SSEs rather than using an eigenvector. Compared with FAST-SSAP and other methods which rely upon the endpoints of SSEs, the simple segregation of secondary structure vectors into six quadrants throws away a lot of more detailed information about the relative orientation of the elements, but as a result, allows us to perform normal single dynamic programming, removing the need for double dynamic programming. Compared with CP, the TOPSCAN representation actually has more information about direction of the secondary structure vectors, but loses the more detailed proximity information (encoded in the constraint programming by hydrogen-bond information) and the chirality arcs. The authors of TOP (Lu, 2000) do not provide directly comparable results. However, since the method relies not only on the direction of the SSEs, but also on the precise location of the endpoints, it seems likely that the method will be particularly sensitive to the secondary structure definition. It is well known that assignment is particularly problematic at the ends of the secondary structures and minor variations between structures can cause one or two residues at each end to be (or not to be) included in an SSE. This could cause a difference in end-point location of >6.0 Å. That is much less of an issue with DEJAVU (Kleywegt and Jones, 1997) the place the matrix of distances between midpoints of the SSEs and angles between the SSEs are used. Whereas the endpoints of the SSEs are utilized by TOPSCAN to outline the path, their exact location is much less necessary. They’re additionally implicitly utilized by TOPSCAN to outline the size of an SSE and of a loop, however since these are merely taken as `lengthy’ or `brief’ somewhat than utilizing an actual size, solely a small share of SSEs or loops (these of size near the cutoff) might be more likely to be misclassified.

The very best outcomes achieved with TOPSCAN (~26.5% protection at 1% error and ~52% protection at 5% error) evaluate with ~60% (at 1% error) and ~78% (at 5% error) for full SSAP and ~36% (at 1% error) and ~58% (at 5% error) for CP (Gilbert et al., 2000). Given the massive enhance in velocity achieved over these strategies, TOPSCAN performs properly and is especially precious as a preliminary display.

Efficiency of TOPSCAN as a prescreen for extra detailed comparability

The protection versus error figures proven in Desk IV give a typical measure for the efficiency of the tactic in construction comparability. As described above, these outcomes are just like the CP technique at a 5% error price.

If the tactic is for use as a pre-screen for a extra detailed and computationally extra demanding technique reminiscent of SSAP, one is extra enthusiastic about with the ability to receive true positives (right hits) with a minimal of false positives (incorrect hits). In abstract, error charges at 90, 95 and 99% protection are 29.02, 39.15 and 60.82%, respectively (it’s impractical to intention for 100% protection owing to odd circumstances reminiscent of area 3 from 1elo and 1dar the place the proteins are similar however have completely different folds in addition to attainable errors in CATH).

Determine 6 exhibits the distribution of the utmost TOPSCAN rating for an accurate hit in opposition to one other protein area of the identical fold. (As earlier than, every protein fold containing a couple of near-identical sequence household was examined and, for every of those households, the utmost TOPSCAN rating for an accurate hit in opposition to one other protein area with the identical fold was recorded.) It’s clear that various right hits have low most scores. Due to this fact, it is strongly recommended that if TOPSCAN is for use as a filter for different strategies, that it’s not used to exclude buildings, however to kind the buildings for presentation to an in depth technique reminiscent of SSAP with the more than likely matches first. As quickly as SSAP finds a great hit, the search via the ordered listing is terminated. This process has been applied utilizing an early model of TOPSCAN (encoding solely secondary construction kind and path) within the CATH-server (Orengo et al., 1998).

Conclusions

TOPSCAN has proved a helpful addition to the accessible strategies for evaluating protein buildings. TOPSCAN was initially developed as a easy technique to scale back the search area for the CATH-server. In that implementation, the secondary topology strings comprise solely the directional data and it’s used as a speedy technique of rating buildings for additional extra detailed comparability utilizing SSAP.

Evaluation of the CATH information utilizing TOPSCAN has proven some fascinating anomalies in addition to a number of minor errors within the CATH information which have been fed again to the CATH maintainers. Thus TOPSCAN additionally seems to have a job in speedy validation of structural classification.

Our evaluation of things affecting the efficiency of TOPSCAN has recommended that treating 310-helix as α-helix in secondary construction prediction is a nasty technique, particularly when the outcomes are for use for fold recognition.

In frequent with different strategies which scale back a three-dimensional construction to a set of secondary construction parts with or with out related vector data, TOPSCAN is more than likely to fail when a construction is poorly outlined and DSSP or STRIDE are unable to make dependable secondary construction assignments. The atomic degree full SSAP is much less susceptible to those issues, however pays an enormous penalty within the time required for comparability.

Future work will embody investigating extra advanced classification of properties reminiscent of solvent accessibility. At present solely binary classifications are used (e.g. uncovered versus buried). Within the case of β-strands, particular allowance could possibly be made for edge strands. Altering the worldwide Needlemann and Wunsch dynamic programming algorithm to a neighborhood Smith–Waterman alignment can even permit for sub-structure matching. Given the present evaluation of things which have an effect on the accuracy of comparability of buildings, this data can feed again into fold recognition and the algorithm described right here, when mixed with strategies for prediction of the elements used within the secondary topology string, might be of software in fold recognition.

A TOPSCAN server has been made accessible at http://www.bioinf.org.uk/topscan/ and at http://www.rubic.rdg.ac.uk/topscan/.

I thank Christine Orengo, Frances Pearl and Janet Thornton for making the CATH information accessible and for his or her help through the first a part of this work, which was funded by departmental funds at College Faculty London as a part of the event of the CATH-server.

 

References – “protein topology”

 

 

“protein topology”