Computation Protein Design instances with small tree-width: selection based on coarse approximated 3D average position volume

David Allouche

REVIEW 4 major objections 5 minor 25 references

Computation Protein Design instances with small tree-width: selection based on coarse approximated 3D average position volume

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A coarse geometric volume metric selects protein redesign instances whose constraint graphs have small tree-width, making global search by decomposition practical.

desk verdict A plausible CPD benchmark with an unverified selection claim: the 21 instances are real but the paper never shows the Vbar filter beats random selection, and the promised supplementary data are missing. read the letter →

arxiv 1909.01803 v1 pith:OTCSTMXX submitted 2019-09-02 cs.CE physics.bio-phq-bio.BM

David Allouche This is my paper · ORCID

classification cs.CEphysics.bio-phq-bio.BM MSC 68T2068R10

keywords computationalproteindesigntree-widthtreedecompositioncostfunctionnetworkradiusofgyrationbenchmarkinstancesredesign

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is a benchmark-construction report: it tries to show that protein redesign problems can be selected for graph decomposability using only a coarse structural number, the average volume per residue estimated from the radius of gyration. Sorting candidate protein structures by this number and keeping 21 extreme examples gives Computational Protein Design instances with 130–282 variables, domain sizes up to 438, and min-fill tree-width 21–68 (normalized 0.16–0.34). These instances are larger than an earlier 50–100 residue benchmark and are offered for testing global search methods that exploit tree decomposition. If the selection rule genuinely produces low tree-width, the set is a harder benchmark for protein design optimization.

What carries the argument

The load-bearing object is the coarse metric $\bar V = (4/3)\pi R_g^3/|X|$, where $R_g$ is the radius of gyration of the protein's atom positions and $|X|$ is the number of designable residue variables. Sorting candidates by decreasing $\bar V$ ranks them from least to most spherical. The link to decomposability is the cutoff used in the energy function: binary cost functions exist only between residues closer than a cutoff, so spherical folds give dense, clique-like constraint graphs with high tree-width, while elongated or irregular folds give sparser graphs with low tree-width. Tree-width, a measure of how close a graph is to a tree, is reported here through the min-fill ordering heuristic.

What would settle it

Compute min-fill tree-width for all 436 candidate proteins, or a random sample of them, and compare the distribution for the 21 selected by $\bar V$ with the distribution for random subsets of equal size; if random subsets routinely have normalized tree-width below 0.34, the geometric filter is not doing the work claimed.

Watch

Extended reading notes

Core claim

The central claim is that the metric $\bar V = (4/3)\pi R_g^3/|X|$, the volume of the gyration-radius sphere divided by the number of designable residues, is a practical proxy for constraint-graph decomposability in computational protein design. Under the paper's protocol—one backbone relaxation and a recent all-atom force field—binary energy terms are added between residue positions closer than a cutoff, so globular proteins produce near-clique graphs while less spherical folds produce sparse, decomposable graphs. Sorting 436 candidate structures by decreasing $\bar V$ and keeping 21 protein chains with at most 40% sequence identity yields min-fill tree-widths between 21 and 68 (0.16 to 0.34 normalized) at sizes of 130–282 variables, larger than previous instances in the same line of benchmarks. The paper also proposes an alternative shape criterion based on gyration-radius components, and notes that both orderings select nearly the same top instances.

Load-bearing premise

The paper assumes, without testing against a baseline, that sorting proteins by coarse average volume per residue picks out redesign problems whose dependency graphs are unusually close to trees.

Editorial extensions

If this is right

The 21 instances, with min-fill tree-width at most 68 and normalized tree-width 0.16–0.34, provide a benchmark on which complete search by tree decomposition is feasible despite redesign sizes of 130–282 residues.
Because the $\bar V$ filter is a simple geometric sort, other groups can build comparable instances from new protein structures without computing graph decompositions in advance.
The near-agreement between the $\bar V$ ordering and the gyration-component criterion suggests that shape anisotropy, not exact volume, is the operative signal for decomposability.
The instances are sized for stress-testing local-search heuristics against exact or decomposition-guided global search, extending an earlier 50–100 residue benchmark to larger problems.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper, the selection rule could be validated directly by computing min-fill tree-width on all 436 candidates; since tree-width is reported only for the 21 chosen proteins, a random-baseline comparison would show whether $\bar V$ actually selects for decomposability or merely tracks protein size and elongation.
A testable extension is to check whether the same $\bar V$ ranking holds under different backbone relaxation protocols or rotamer libraries, because the cutoff-dependent edge set of the constraint graph changes with the energy function.
Small normalized tree-width does not by itself guarantee easy optimization: cost landscapes can be hard even on tree-shaped graphs, so future comparisons should separate decomposability from search difficulty.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague, the thing to know about this paper is that it is a benchmark proposal whose central selection claim is unverified as posted. The 21 CPD instances are real and somewhat larger than the earlier Simoncini et al. set, but the paper never shows that the Vbar ordering actually yields low tree-width relative to the other 415 candidates from the same PDB query, and the promised companion spreadsheet with the full PDB list is not attached. So the headline claim "small tree-width" describes the selected table, not a demonstrated selection effect.

What is good: the instances are built with a newer Rosetta energy function (BetaNov16), one backbone relaxation, and sizes from 130 to 282 variables with domain sizes up to 438. If you work on tree-decomposition-based CPD search, this is a plausible stress set. The Vbar metric (coarse gyration volume per variable) is simple and cheap, and the paper also checks an alternative Rg-component criterion that mostly reproduces the same ordering for the first 18 proteins, which is a useful sanity check. The paper is honest in Sections 5.2 and 6 that these are heuristics rather than exact characterizations.

The soft spots are load-bearing. There is no baseline comparison: no tree-width distribution for the 436-candidate pool, no random-selection comparison, and no comparison to the older Simoncini protocol. Without that, the claim that the filter selects unusually decomposable instances is not supported. Relatedly, the reported min-fill tree-width is itself a heuristic upper bound, not exact tree-width, and the abstract and conclusion do not say so. The physical intuition in Section 5.2 is plausible but not automatic: a distance cutoff does not turn a globular protein into a clique, and an elongated protein does not guarantee a small tree-width contact graph. The missing supplementary spreadsheet matters because a benchmark paper is only useful if the instances are available. Those are fixable, but they are not cosmetic.

Would I send this to peer review? Yes, as a short benchmark note with required revisions: add the data, add the baseline, and qualify the tree-width claim. It does not deserve a desk reject, but it should not be accepted as-is. I would not cite it in its current form.

Referee Report

4 major / 5 minor

Summary. The manuscript describes a protocol for selecting 21 computational protein design (CPD) benchmark instances from a PDB query of 436 candidate proteins. Selection uses a coarse structural filter V̄ = (4/3)πRg³/|X| defined in Eq. (3), and the paper reports min-fill tree-width values between 21 and 68 (normalized 0.16–0.34) in Table 1. The stated goal is to provide larger instances (130–282 variables) than the earlier Simoncini et al. benchmark, with graph-decomposition-friendly constraint graphs, for evaluating solvers such as UDGVNS.

Significance. If the selected instances truly have unusually small tree-width relative to the candidate pool, they would constitute a useful benchmark for testing decomposition-based and hybrid search methods on CPD instances larger than earlier 50–100 residue sets. The paper usefully identifies concrete PDB structures and reports per-instance statistics, and it uses a standard Rosetta energy model. However, the central selection claim is not currently validated: no baseline comparison is provided, and the promised supplementary data are missing. The added value of the V̄ filter over random selection therefore remains unquantified.

major comments (4)

[§5.2, Table 1] The central claim that the V̄ filter selects instances with small tree-width is not supported by any baseline comparison. Table 1 lists min-fill tree-width for the 21 selected proteins only; the paper does not report tree-width statistics for the remaining 415 candidates from the same PDB query, for a random selection from that pool, or for the Simoncini et al. protocol. Because the selection criterion is the paper's main contribution, the authors should provide this distribution and show that the selected instances lie at the favorable extreme; without it, the reader cannot determine whether the filter adds value beyond arbitrary selection.
[§5.2, Eq. (3)] The assertion that globular proteins 'will closely correspond to clique' and that non-spherical proteins 'induce constraint graph sparsity' is an unproved intuition rather than a derivation. The CFN graph is defined by a distance cutoff on side-chain positions, so a globular 100–300 residue protein is not a clique, and an elongated or non-spherical shape does not guarantee low treewidth: a thick tubular contact graph can have treewidth growing with its cross-section. If the V̄ heuristic is to be credible, the authors should either prove a quantitative relation between shape and treewidth or provide an empirical comparison of edge density and treewidth over the candidate pool; the current 'by construction' wording overstates the case.
[§5.1, §6] The manuscript promises a supplementary spreadsheet with the full PDB identifier list and detailed instance information, but no such attachment appears in the submitted arXiv package. Without this data, the 21 instances cannot be regenerated, verified, or reused by other groups. The authors must include the spreadsheet or a stable repository with the exact instance files, generation parameters, and the complete 436-entry candidate list.
[Table 1] The values reported as tree-width are min-fill tree-widths, which are heuristic upper bounds rather than exact tree-widths. Since the benchmark's stated purpose is to provide small-tree-width instances for decomposition-based methods, the paper should either compute exact tree-width (or at least a tighter certified bound) or consistently describe the reported values as upper bounds; the normalized ratio tw/|X| inherits this caveat.

minor comments (5)

[Figure 2 caption] The caption contains multiple typos, including 'non Gobular protein will induce constraint graph sparcity' and 'spherical protein will be clause from click'; the last phrase should likely be 'close to a clique'.
[§5.2 heading] The heading '3D critter filtering' should be '3D criteria filtering'.
[§5.3] The claim that the alternative criterion in Eq. (7) produces the same 18 first instances with re-ranking is not supported by any table or data; please provide the comparison.
[Abstract and references] The abstract uses 'Simononcini' while the body and reference [2] use 'Simoncini'; please harmonize the spelling throughout.
[Equations] Equation references such as '(eq: 3)' and '(eq: 7)' appear as unformatted text; please use proper cross-references.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the geometric filter and tree-width measurements are independent; the central claim is under-validated but not derived from its inputs.

full rationale

The paper's derivation chain is: generate CFN instances following the protocol of [2] with Rosetta; compute a geometric heuristic V_bar = (4/3)π Rg^3 / |X| from PDB coordinates; sort the 436 candidates by decreasing V_bar; take the first 21; then measure min-fill tree-width on the resulting CFN constraint graphs and report them in Table 1. The tree-width values are not used to choose the 21 proteins: the selection is made purely from V_bar before any tree-width computation, so the reported tw numbers are independent measurements, not a renamed or fitted input. The paper explicitly labels V_bar a 'heuristic' (Section 5 and Conclusion: 'both of which are heuristic') and says the goal is to detect non-globular characters 'due to their putative highly decomposable properties', so it does not claim a first-principles derivation of tree-width from the filter. There is no fitted parameter calibrated on tree-width data, no uniqueness theorem imported from the authors' prior work, and no equation that reduces to another by construction. The self-citations to [1], [2], [4], [11], [20] and [25] are methodological or contextual (CFN modeling, instance-generation protocol, VNS algorithm) and are not load-bearing for the selection claim. The paper's weaknesses are missing validation—no comparison with the 415 non-selected candidates—and a promised but absent supplementary spreadsheet (Section 5.1: 'The full list of the corresponding pdb identiﬁers is available in the supplementary spreadsheet document'), but those are empirical-support and completeness gaps, not circular reasoning. Therefore the circularity score is 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The paper's central claim rests on a geometric heuristic whose predictive power is assumed, not demonstrated. No new entities or fitted parameters are introduced; the selection cutoff of 21 proteins is an arbitrary hand choice.

free parameters (1)

number of selected proteins = 21
The benchmark size is an arbitrary hand-chosen cutoff from the 436 candidate proteins; the paper gives no stopping rule or justification for why 21 is the right number.

assumptions (3)

domain assumption Globular proteins produce CFN constraint graphs close to cliques, while non-spherical proteins produce sparse, well-decomposable graphs.
Invoked in Section 5.2 to justify the V_bar ordering, but no quantitative evidence or empirical correlation is provided.
domain assumption The min-fill heuristic tree-width approximates the cost function network's actual tree-width.
Table 1 reports min-fill widths as 'tw', and the claim of small tree-width depends on this number being representative.
domain assumption The BetaNov16 Rosetta force field and Dunbrack rotamer library correctly define the CPD energy model.
The instances are built on these external models, cited as [3] and [17], and are treated as ground truth without verification in this paper.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Computation Protein Design instances with small tree-width: selection based on coarse approximated 3D average position volume." pith.science (2026). https://pith.science/paper/OTCSTMXX

@misc{pith2026190901803,
  author       = {Pith},
  title        = {Pith review of: Computation Protein Design instances with small tree-width: selection based on coarse approximated 3D average position volume},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OTCSTMXX}},
  note         = {Machine review of arXiv:1909.01803}
}

read the original abstract

This paper proposes small tree-width graph decomposition computational protein design CFN instances defined according to the model [1] with protocol defined by Simononcini et al [2] . The proteins used in the benchmark have been selected in the PDB (not on their biological interest) to explore the efficiency of global search method based on tree-width decomposition. The instances are bigger than those previously proposed in the paper [2] with one backbone relaxation and the aka Beta November 2016 Rosetta force-field [3]. The benchmark includes 21 proteins selected with a low level of sequences identity (40%) . Those instances have been selected on the basis of 3D criteria by applying a decreasing average coarse volume occupancy filter by Amino Acid (-i.e. by CFN variable) . The instances characteristic (see Table 1) contain from 130 up to n = 282 variables with a maximum domain size from 383 to 438, and between 1706 and 6208 cost functions. The min-fill tree-width ranges from 21 to 68, and from 0.16 to 0.34 for a normalized tree width. Those instances have been used for UDGVNS search algorithm[4] benchmarking. This approach is suitable for evaluation of search methods that exploit the notion of graph decomposition.

Figures

Figures reproduced from arXiv: 1909.01803 by the authors.

**Figure 2.** schematic view of cfn modeling for single backbone protein design: [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** A representation of distribution of the number of residues by entries in [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: A look at the diversity of structures in the PDB archive. These images, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: ryration radius representation on a given pdb structure [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 24 canonical work pages

[1]

Allouche, et al., Computational protein design as a cost function network optimization problem, in: Proc

D. Allouche, et al., Computational protein design as a cost function network optimization problem, in: Proc. of CP, 2012, pp. 840–849. 10

work page 2012
[2]

Simoncini, D

D. Simoncini, D. Allouche, S. de Givry, C. Delmas, S. Barbe, T. Schiex, Guaranteed discrete energy optimization on large protein design problems, J. of Chemical Theo. and Comput. 11(12) (2015) 5980–5989

work page 2015
[3]

Alford, et al., The rosetta all-atom energy function for macromolecular modeling and design, Journal of Chemical Theory and Computation 13 (6) (2017) 3031–3048

R. Alford, et al., The rosetta all-atom energy function for macromolecular modeling and design, Journal of Chemical Theory and Computation 13 (6) (2017) 3031–3048

work page 2017
[4]

Ouali, D

A. Ouali, D. Allouche, S. de Givry, S. Loudni, Y . Lebbah, F. Eckhardt, L. Loukil, Iterative Decomposition Guided Variable Neighborhood Search for Graphical Model Energy Minimization, in: Proc. of UAI-17, Sydney, Australia, 2017, pp. 550–559

work page 2017
[5]

S. M. Lippow, B. Tidor, Progress in computational protein design, Current Opinion in Biotechnology 18 (4) (2007) 305 – 311, protein technologies / Systems biology. doi:https://doi.org/10.1016/j.copbio. 2007.04.009. URL http://www.sciencedirect.com/science/article/ pii/S0958166907000778

doi:10.1016/j.copbio 2007
[6]

Huang, S

P.-S. Huang, S. E. Boyken, D. Baker, The coming of age of de novo protein design, Nature 537 (2016) 320–327

work page 2016
[7]

Baker, What has de novo protein design taught us about protein folding and biophysics?, Protein Science 28 (4) (2019) 678–683

D. Baker, What has de novo protein design taught us about protein folding and biophysics?, Protein Science 28 (4) (2019) 678–683. arXiv:https: //onlinelibrary.wiley.com/doi/pdf/10.1002/pro.3588, doi:10.1002/pro.3588. URL https://onlinelibrary.wiley.com/doi/abs/10. 1002/pro.3588

work page doi:10.1002/pro.3588 2019
[8]

Pabo, Molecular technology

C. Pabo, Molecular technology. Designing proteins and peptides., Nature 301 (5897) (1983) 200

work page 1983

Show all 25 references

[9]

B. I. Dahiyat, S. L. Mayo, Protein design automation., Protein science 5 (5) (1996) 895–903

1996
[10]

N. A. Pierce, E. Winfree, Protein design is NP-hard., Protein engineering 15 (10) (2002) 779–82

2002
[11]

Allouche, I

D. Allouche, I. Andr ´e, S. Barbe, J. Davies, S. de Givry, G. Katsirelos, B. O’Sullivan, S. Prestwich, T. Schiex, S. Traor´e, Computational protein de- sign as an optimization problem, Artiﬁcial Intelligence 212 (2014) 59–79

2014
[12]

G. A. Khoury, J. Smadbeck, C. A. Kieslich, C. A. Floudas, Protein fold- ing and de novo protein design for biotechnological applications, Trends in biotechnology 32 (2) (2014) 99–109. 11

2014
[13]

Gront, D

D. Gront, D. W. Kulp, R. M. Vernon, C. E. Strauss, D. Baker, Generalized fragment picking in rosetta: design, protocols and applications, PloS one 6 (8) (2011) e23294

2011
[14]

J. U. Bowie, R. Luthy, D. Eisenberg, A method to identify protein sequences that fold into a known three-dimensional structure, Science 253 (5016) (1991) 164–170

1991
[15]

Anﬁnsen, Principles that govern the folding of protein chains, Science 181 (4096) (1973) 223–253

C. Anﬁnsen, Principles that govern the folding of protein chains, Science 181 (4096) (1973) 223–253

1973
[16]

Janin, S

J. Janin, S. Wodak, M. Levitt, B. Maigret, Conformation of amino acid side- chains in proteins, Journal of molecular biology 125 (3) (1978) 357–386

1978
[17]

Campeotto, A

F. Campeotto, A. Dal Pal, A. Dovier, F. Fioretto, E. Pontelli, A constraint solver for ﬂexible protein models, J. Artif. Int. Res. (JAIR) 48 (1) (2013) 953–1000

2013
[18]

S. C. Lovell, J. M. Word, J. S. Richardson, D. C. Richardson, The penultimate rotamer library., Proteins 40 (3) (2000) 389–408

2000
[19]

Tuffery, C

P. Tuffery, C. Etchebest, S. Hazout, R. Lavery, A new approach to the rapid determination of protein side chain conformations., Journal of biomolecular structure & dynamics 8 (6) (1991) 1267–89

1991
[20]

Charpentier, D

A. Charpentier, D. Mignon, S. Barbe, J. Cortes, T. Schiex, T. Simonson, D. Allouche, Variable neighborhood search with cost function networks to solve large computational protein design problems, Journal of Chemical In- formation and Modeling 59 (1) (2019) 127–136

2019
[21]

H. M. Berman, J. Westbrook, Z. Feng, G. Gilliland, T. Bhat, H. Weissig, I. N. Shindyalov, P. E. Bourne, The protein data bank, Nucleic acids research 28 (1) (2000) 235–242

2000
[22]

H. M. Berman, The Protein Data Bank: a historical perspective, Acta Crystallographica Section A 64 (1) (2008) 88–95. doi:10.1107/ S0108767307035623. URL https://doi.org/10.1107/S0108767307035623

2008 doi
[23]

Leaver-Fay, M

A. Leaver-Fay, M. Tyka, S. M. Lewis, O. F. Lange, J. Thompson, R. Ja- cak, K. Kaufman, P. D. Renfrew, C. A. Smith, W. Shefﬂer, I. W. Davis, S. Cooper, A. Treuille, D. J. Mandell, F. Richter, Y .-E. A. Ban, S. J. Fleish- man, J. E. Corn, D. E. Kim, S. Lyskov, M. Berrondo, S. Me...

2011
[24]

M. J. O’Meara, A. Leaver-Fay, M. Tyka, A. Stein, K. Houlihan, F. DiMaio, P. Bradley, T. Kortemme, D. Baker, J. Snoeyink, B. Kuhlman, A combined covalent-electrostatic model of hydrogen bonding improves structure predic- tion with rosetta, J. Chem. Theory Comput. 11 (2) (2015) 609–622

2015
[25]

Ouali, D

A. Ouali, D. Allouche, S. De Givry, S. Loudni, Y . Lebbah, F. Eckhardt, L. Loukil, Iterative decomposition guided variable neighborhood search for graphical model energy minimization, in: Conference on Uncertainty in Ar- tiﬁcial Intelligence, UAI 2017, 2017. 13

2017

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Allouche, et al., Computational protein design as a cost function network optimization problem, in: Proc

D. Allouche, et al., Computational protein design as a cost function network optimization problem, in: Proc. of CP, 2012, pp. 840–849. 10

work page 2012

[2] [2]

Simoncini, D

D. Simoncini, D. Allouche, S. de Givry, C. Delmas, S. Barbe, T. Schiex, Guaranteed discrete energy optimization on large protein design problems, J. of Chemical Theo. and Comput. 11(12) (2015) 5980–5989

work page 2015

[3] [3]

Alford, et al., The rosetta all-atom energy function for macromolecular modeling and design, Journal of Chemical Theory and Computation 13 (6) (2017) 3031–3048

R. Alford, et al., The rosetta all-atom energy function for macromolecular modeling and design, Journal of Chemical Theory and Computation 13 (6) (2017) 3031–3048

work page 2017

[4] [4]

Ouali, D

A. Ouali, D. Allouche, S. de Givry, S. Loudni, Y . Lebbah, F. Eckhardt, L. Loukil, Iterative Decomposition Guided Variable Neighborhood Search for Graphical Model Energy Minimization, in: Proc. of UAI-17, Sydney, Australia, 2017, pp. 550–559

work page 2017

[5] [5]

S. M. Lippow, B. Tidor, Progress in computational protein design, Current Opinion in Biotechnology 18 (4) (2007) 305 – 311, protein technologies / Systems biology. doi:https://doi.org/10.1016/j.copbio. 2007.04.009. URL http://www.sciencedirect.com/science/article/ pii/S0958166907000778

doi:10.1016/j.copbio 2007

[6] [6]

Huang, S

P.-S. Huang, S. E. Boyken, D. Baker, The coming of age of de novo protein design, Nature 537 (2016) 320–327

work page 2016

[7] [7]

Baker, What has de novo protein design taught us about protein folding and biophysics?, Protein Science 28 (4) (2019) 678–683

D. Baker, What has de novo protein design taught us about protein folding and biophysics?, Protein Science 28 (4) (2019) 678–683. arXiv:https: //onlinelibrary.wiley.com/doi/pdf/10.1002/pro.3588, doi:10.1002/pro.3588. URL https://onlinelibrary.wiley.com/doi/abs/10. 1002/pro.3588

work page doi:10.1002/pro.3588 2019

[8] [8]

Pabo, Molecular technology

C. Pabo, Molecular technology. Designing proteins and peptides., Nature 301 (5897) (1983) 200

work page 1983