Skip to content
Kosha Sciences
← Research
RCCProtein design

Recursive Constraint Compilation for Structured Backbone Design

Protein backbone design with geometric constraints built in

An antibody recognizes its target through a surface shaped in part by flexible protein loops. Redesigning one of those loops offers a way to change that interaction while keeping the surrounding framework. But the new segment has to fit: both ends must meet the existing structure, and the chain between them must obey its geometric constraints.

That requirement can conflict with the way a generative model searches. Moving a backbone toward favorable contacts can pull an endpoint away from its anchor or distort the spacing between residues. Repairing the geometry afterward can remove the very contacts that made a candidate look promising. The design we score and the design we can retain are then different structures.

Recursive Constraint Compilation (RCC) builds fixed endpoints and link lengths into the coordinates used for generation and optimization. Every candidate is decoded into a backbone that satisfies those constraints before it is scored. The search can then compare feasible shapes, while binding, sterics, and other aspects of protein quality remain separate questions.

RCC paper overview: a generated backbone is normalized and encoded through a tree, then decoded with the same coordinates under the required endpoint displacement, enforcing endpoints and link lengths.View full-size figure

Compile the constraint into the coordinates. The original paper’s overview follows a sample from a pretrained generator into recursive coordinates and back into a fixed-endpoint backbone. The same construction supplies the volume correction needed for probability densities, and coordinates that support feasible guidance.

Figure 1 in the RCC paper

Define the valid design space

Local protein design appears in several forms: replacing an antibody loop, filling a missing fragment, or connecting functional motifs. In each case, some structure is fixed and some is free to change. The boundary between them is a geometric requirement, not just a preference in the loss function.

The structure around the design

Structure study
Neuraminidase · DHeavy chain · ELight chain · I

The native complex used as context for the paper’s Figure 3 (8yvn E I D). Space-filling atoms show the antibody–antigen interface; smoothed Cα traces reveal the three chains. Both views use the deposited cryo-EM atomic model and are fitted separately.

PDB 8YVN · cryo-EM atomic model ↗

RCC represents a backbone segment by its Cα positions. A link is the displacement between neighboring positions: xi=pi−pi−1x_i=p_i-p_{i-1}. With NN links, prescribed length bb, and fixed endpoint displacement ee, the allowed configurations form a bridge space:

The fixed-endpoint bridge(1)
BN(e)={(x1,…,xN): ∥xi∥=b, ∑i=1Nxi=e},dim⁡BN(e)=N(d−1)−d.\begin{aligned}\mathcal B_N(e)&=\left\{(x_1,\ldots,x_N):\ \lVert x_i\rVert=b,\ \sum_{i=1}^{N}x_i=e\right\},\\[6pt]\dim\mathcal B_N(e)&=N(d-1)-d.\end{aligned}
Each link lies on a sphere of radius b, while their sum fixes the endpoint. At noncollinear configurations, a three-dimensional bridge has 2N − 3 continuous degrees of freedom.

This equation separates hard geometry from the rest of the design problem. A model may still need to learn favorable conformations, and a design objective may still need to reward contacts or penalize clashes. RCC makes the endpoints and Cα spacing exact on its regular coordinate domains. It does not make those other properties automatic.

The construction assumes a reachable endpoint, ∥e∥<Nb\lVert e\rVert<Nb. The paper develops equal-length links. Its central question is how to parameterize this space so that generation, density evaluation, and gradient-based search can all use it. [§2.2]

Turn one global constraint into compatible local choices

Imagine cutting a chain into a left half and a right half. Once their meeting point is chosen, each half becomes a smaller fixed-endpoint problem. The whole chain can therefore be constructed by repeatedly making compatible splits.

For a parent segment from zero to DD, let the two children contain nℓn_\ell and nrn_r links. The left child can reach a meeting point XX only if it lies within distance nℓbn_\ell b of the start. The right child must also be able to reach the endpoint from that same point. Their reachable regions overlap in a lens.

The meeting point must work for both children(2)
L(D)=B(0,nℓb)∩B(D,nrb),X=c+rρD(ω) ω,c=nℓnℓ+nrD.\begin{aligned}\mathcal L(D)&=B(0,n_\ell b)\cap B(D,n_r b),\\[6pt]X&=c+r\rho_D(\omega)\,\omega,\qquad c=\frac{n_\ell}{n_\ell+n_r}D.\end{aligned}
The unit direction ω and radial fraction 0 < r < 1 select a point inside the lens. The radius ρ is the distance from c to the lens boundary in that direction. The point c is a geometric center chosen by the construction, not the protein context used later.

The coordinate domain stays the same when the endpoint displacement changes: choose a direction and a fraction of the available radius. The decoder accounts for the changing shape of the lens. Applying this rule to both children eventually reaches two- and three-link chains with explicit solutions.

Building an antibody loop

CAV-F6 · PDB 8YVN
RCC geometry demonstration
Loading antibody structure…
CDR-H316 residues · two fixed anchors
Constructed H3 backboneAntibody frameworkNeuraminidase
  1. 01Fixed anchors
  2. 02First split
  3. 03Recursive splits
  4. 04Completed loop
Static projection while 3D loadsExample 01 / 03 · Cα spacing 3.8 Å

RCC constructs a backbone for the IMGT-defined H3 loop between anchors E96 and E113 in the deposited CAV-F6 complex. Dashed spans represent subchains waiting to be resolved; the final 17 Cα links are 3.8 Å long. These independently generated geometry examples use a simple clash-and-bend screen. The paper’s trained sampler was not used, and the candidates have not been validated as antibody designs.

Structure and loop annotation ↗
Explore the geometry of a single split

Choose a split. Keep the bridge closed.

INTERACTIVE · 3D GEOMETRY
Fixed startFixed endMeeting point X
Loading interactive 3D view…
Four left linksFour right linksReachable volume
Max link error: 6.7e-16Endpoint error: 0.0e+0
One recursive split in three dimensions. Rotate the view and move the sliders to explore the chain. The translucent lens is the overlap of two reachable spheres. The depth angle turns the split around the fixed endpoint axis; the radius sets its fraction of the distance to the lens boundary. Every meeting point remains inside that volume, and all eight links retain unit length. Child split coordinates and leaf-circle angles are fixed for this mathematical illustration. These are illustrative coordinates, not a sampled protein. Errors are calculated from the displayed geometry.

The general RCC framework asks each subtree to expose a compact summary. For a backbone, that summary is its endpoint displacement. The parent displacement is the sum of the two child displacements. If every local compatibility condition and every leaf has invertible coordinates, these maps assemble into a global encoder EeE_e and decoder DeD_e.

On the chosen regular domains, decoding always satisfies the bridge constraint, and encoding and decoding undo one another. A balanced tree gives linear total work and logarithmic parallel depth for fixed spatial dimension. This is a statement about the recursive construction, not a claim that the entire protein-design pipeline has logarithmic runtime. [Theorem 2.2]

Where the coordinate guarantee applies

The maps are exact and invertible on compatible regular charts. They are not one smooth global chart covering every possible bridge. A lens-center split has no unique radial direction; collinear subchains and boundary cases can also be singular. Appendix A.2 gives an example where a noncollinear overall bridge still encounters a singular intermediate split. The numerical checks and density formulas retain these regularity conditions.

Connect the coordinates to existing generators

A useful constraint representation should work with an existing generator. RCC supplies a retraction: a map that takes a generated backbone into the required bridge space and leaves an already-valid backbone unchanged, whenever the relevant normalization and charts are well defined.

For generated links yiy_i, the first step distributes endpoint mismatch across the chain and normalizes every link to length b. This produces an equal-link source bridge q with its own endpoint displacement eye_y. RCC encodes that bridge, then decodes the same recursive coordinates under the required endpoint e.

Keep the coordinates; change the endpoint(3)
δ=e−∑jyjN,qi=byi+δ∥yi+δ∥,ey=∑iqi,Re(y)=De ⁣(Eey(q)).\begin{aligned}\delta&=\frac{e-\sum_j y_j}{N},\qquad q_i=b\frac{y_i+\delta}{\lVert y_i+\delta\rVert},\\[6pt]e_y&=\sum_i q_i,\qquad R_e(y)=D_e\!\left(E_{e_y}(q)\right).\end{aligned}
Normalization establishes the source bridge. Encoding and decoding transfer its recursive choices to the target endpoint. For a valid input x in the regular domain, Rₑ(x) = x.

This operation does not require retraining the underlying generator or changing its sampling procedure. It also does not minimize Cartesian displacement from the original sample: preserving recursive coordinates is a different choice from finding the nearest feasible backbone. [§2.3]

There are two distinct uses of the same geometry. The retraction above can close a sample after generation. During guided design, RCC can instead decode every candidate before it is evaluated. The second use is what prevents the optimizer from exploiting invalid geometry.

A coordinate density is not yet a backbone density

Equal-sized neighborhoods in coordinate space need not occupy equal amounts of backbone space. A decoder can stretch one region more than another. A normalized probability density over backbones therefore needs a volume correction.

Density with respect to intrinsic backbone volume(4)
qB(x∣c,e)=qψ(Ee(x)∣c,e)Je(Ee(x))q_{\mathcal B}(x\mid c,e)=\frac{q_\psi(E_e(x)\mid c,e)}{J_e(E_e(x))}
Here c denotes the visible protein context, qψ is a normalized coordinate density, and Jₑ is the decoder’s intrinsic-volume factor. The formula applies on a one-to-one regular chart, with zero density outside its image.

RCC computes this factor through the same tree. Each node combines its children’s volume factors with the local meeting-point Jacobian and the amount by which both children deform when that point moves. This avoids constructing the full backbone decoder Jacobian.

The recursive volume calculation

For a subtree I, each matrix term projects onto directions in which a link can rotate without changing its length. Their sum describes how those rotations can move the subtree endpoint. At a regular split, the two children contribute jointly:

Combine child volumes and endpoint motion(5)
MI=∑i∈I(Id−x^ix^iT),x^i=xi/b,JI=JℓJr ∣det⁡Bv∣ det⁡ ⁣(Mℓ−1+Mr−1),MI=Mℓ+Mr.\begin{aligned}M_I&=\sum_{i\in I}\left(I_d-\hat x_i\hat x_i^{\mathsf T}\right),\qquad \hat x_i=x_i/b,\\[8pt]J_I&=J_\ell J_r\,\lvert\det B_v\rvert\,\sqrt{\det\!\left(M_\ell^{-1}+M_r^{-1}\right)},\\[6pt]M_I&=M_\ell+M_r.\end{aligned}
Bᵥ is the Jacobian of the local meeting point with respect to its split coordinates. The recurrence assumes noncollinear children and a nonsingular local Jacobian. Two- and three-link formulas initialize the leaves.

The total work is O(Nd³), with O(d³ log N) parallel depth. For proteins, d = 3 is fixed. This factor is needed for normalized densities; a geometry-only guidance implementation can omit it when the loss does not use density. [Theorem 2.4]

The paper checks these identities independently. Across 240 prespecified protein-bridge configurations with 2–31 links, the maximum float64 constraint residual is 4.07 × 10−15. Recursive log volume agrees with the full decoder Jacobian to 1.42 × 10−14, and coordinate gradients to 1.91 × 10−14. These checks establish numerical agreement on tested regular domains, rather than empirical protein quality. [Table 6]

Score the backbone that can actually be built

A learned sampler maps base noise ε into coordinates u using TψT_\psi. The backbone decoder turns those coordinates into a valid chain. Guidance changes the base noise to improve a differentiable design objective:

Optimize through the feasible decoder(6)
min⁡ε  L ⁣(De ⁣(Tψ(ε;c,e)),c)\min_{\varepsilon}\;\mathcal L\!\left(D_e\!\left(T_\psi(\varepsilon;c,e)\right),c\right)
Every scored backbone is decoded under the required endpoint and link lengths. The hard constraints come from Dₑ; the design loss decides which feasible backbones are preferred.
  1. Initialize. Draw four candidates from the learned conditional proposal.
  2. Decode and score. Construct a valid backbone for each candidate before evaluating its contacts, sterics, and local geometry.
  3. Update. Take 15 Adam steps through the sampler and decoder, retaining the best objective encountered.

Scoring the initial candidates and every update gives 4 × (15 + 1) = 64 feasible objective evaluations per condition. The prospective study withholds the native H3 sequence and coordinates from the model, design objective, ranking, and success criteria. Native RMSD is reserved for retrospective analysis. [Algorithm 2]

What the experiments show

The first evaluation asks how much closing existing generator outputs changes their structural summaries. RCC is applied to 20,610 LoopGen CDR samples, 212,940 CATH fragment samples, and 292 FrameFlow motif-linker samples. All 233,842 outputs satisfy the reported geometric feasibility criterion.

Compared with fixed-512 projection on the same inputs, changes in each setting’s primary structural measure are below 0.01 Å. That is a comparison between closure procedures, not a claim that each backbone moves by less than 0.01 Å from its raw generated state. On CATH, RCC uses one fixed encode–decode pass; FABRIK has the lowest reported p99 latency. [Tables 1–2]

Prospective antibody-loop design

The main guidance study uses a PDB-disjoint temporal panel of antibody–antigen complexes. The checkpoint and evaluation protocol were fixed before the temporal test structures were accessed. Of 292 initial conditions, four fail the prespecified eligibility criteria, leaving 288 conditions from 255 PDBs, with H3 lengths of 6–26 residues.

Design conditions
288
Parent PDBs
255
Evaluations / condition
64

The learned proposal assigns native loops higher density than uniform coordinates in 92.0% of conditions. Permuting context across PDBs lowers mean native-loop log density by 1.152 nats, with the matched context preferred in 61.3% of conditions. The model is using information about the surrounding antibody and epitope, beyond a length-and-endpoint description alone.

Learning helps. Feasible guidance adds more.

PROSPECTIVE · TABLE 3
Joint structural success: uniform sampling 2.78%, learned sampling 42.01%, RCC guidance 70.14%.
A matched inference budget. All three policies receive 64 feasible objective evaluations for each of 288 conditions from 255 PDBs. The percentages measure the paper’s joint structural criteria, not binding or experimental success. The chart shows reported aggregate values; no per-bar confidence intervals are supplied in Table 3. Source: Table 3.
See the structural metrics
PolicyObjective ↓Cα clean ↑Diversity (Å) ↑Native RMSD (Å) ↓
Uniform sampling0.5915628.13%11.1029.063
Learned sampling0.1070381.60%5.3435.564
RCC guidance0.0165398.61%5.8416.008

Learning the proposal raises structural success from 2.78% to 42.01%; coordinate guidance raises it to 70.14%, an increase of about 28 percentage points over learned sampling. Every one of the 18,432 backbones evaluated during guidance satisfies the endpoint and link-length constraints. The objective improves over learned sampling on 278 of 288 conditions. [Appendix E.1]

The result is not native-loop reconstruction. Mean native Cα RMSD rises from 5.564 to 6.008 Å under guidance, while diversity rises from 5.343 to 5.841 Å. The objective favors the paper’s context-directed structural criteria rather than similarity to the withheld native H3.

Why optimizing first and repairing later falls short

A useful control keeps the starting candidates and scoring budget fixed, then changes the space in which optimization runs. The paper repeats the same four-start, 15-step search in Cartesian link coordinates and repairs the final candidates afterward.

Before repair, the Cartesian search reaches a lower mean objective than RCC guidance: 0.00755 versus 0.01653. But those candidates violate the bridge constraints, with a maximum normalized link error of 3.41. Their apparently better score belongs to geometry that the final design is not allowed to have.

Repair restores geometry, but loses the score.

POST HOC · TABLE 20
Mean objective is 0.01653 for RCC guidance, versus 0.54716 to 0.69009 for Cartesian guidance followed by four different repair methods. Lower is better.
Every output shown is feasible. The repaired arms share the same four initial samples and 64 objective evaluations with RCC guidance. Before repair, the invalid Cartesian candidates score 0.00755; after repair, their mean objective is much worse. Budgets match objective calls, not backward-pass cost or elapsed time. Source: Table 20.
Compare success after each repair method
Search and terminal mapStructural success ↑Cα clean ↑
RCC-coordinate guidance70.14%98.61%
Cartesian + adaptive projection7.64%48.61%
Cartesian + FABRIK6.60%45.49%
Cartesian + CoBarS6.94%44.79%
Cartesian + RCC retraction11.11%53.47%

All four terminal methods restore valid geometry. None preserves the objective improvement: structural success after repair ranges from 6.60% to 11.11%, compared with 70.14% for direct coordinate guidance. Even the RCC retraction performs poorly when used only at the end. The distinction is maintaining feasibility during search, not simply choosing RCC as the final correction.

Original paper Figure 3: antibody context and native H3, RCC-guided backbone, and a Cartesian candidate changed by adaptive repair.
The effect on an actual backbone. The paper’s Figure 3 shows condition 8yvn E I D. Left: the design context and native H3, withheld during inference. Center: the initial and RCC-guided candidates. Right: a Cartesian optimum and its adaptively repaired backbone. This example was selected after the study for visualization and is not an additional statistical test. View original figure ↗

This control is post hoc. It matches initial samples and objective-evaluation count, not total computation. It also does not establish superiority over every method that keeps candidates feasible throughout optimization. That comparison is a separate question.

Exact geometry leaves important design questions open

The stronger feasibility-maintaining controls make the conclusion more precise. In a separate development comparison, RCC-coordinate guidance performs comparably to Riemannian guidance and Adam differentiated through closure. Across three proposal-training seeds, the RCC-minus-Riemannian difference in the combined structural-and-steric criterion is −0.95 percentage points, with a descriptive 95% interval of −7.31 to +5.39 points. The paper does not establish a general optimization advantage over these methods. [Appendix E.5]

Backbone resolution also matters. The original guidance objective improves Cα cleanliness, yet the reconstructed atom-level clean fraction is lower than for learned sampling. A later atom-aware objective changes that tradeoff:

Cα cleanliness is not atom-level cleanliness.

POST-HOC CHECK · TABLE 23
PolicyCα clean ↑Reconstructed atom clean ↑
Learned samplingProspective outputs81.60%23.26%
RCC guidanceProspective outputs98.61%19.79%
Atom-aware RCC guidancePost hoc100.00%75.00%
The atom-level criterion is evaluated after PULCHRA reconstruction. Its clean fraction falls under the original guidance objective, even though Cα cleanliness rises. A later atom-aware objective improves reconstructed sterics, with structural success of 65.28%. All reconstruction checks are post hoc; neither criterion establishes folding or binding. Source: Table 23.

Independent sequence-design and refolding checks are more demanding still. In the first development pilot, none of the 64 sequence attempts for RCC-coordinate Adam comes within 2 Å of its designed H3 backbone after refolding. The other guided methods also have zero passing attempts. Exact closure and good coarse structural scores do not establish sequence–structure self-consistency. [Table 26]

A later, distinct endpoint-conditioned generation experiment reports 24 of 128 sequence attempts passing the same 2 Å criterion with synchronized atoms, across 11 of 32 PDBs. These are development results from a different pipeline, not an update to the prospective 70.14% figure. The native-sequence controls perform substantially better, training-set overlap with the refolding model is unresolved, and none of these experiments measures physical folding or binding. [Table 28]

A reusable construction for constrained design

The recursive idea is broader than chain geometry. The paper also considers allocating a fixed budget among bounded variables. Each subtree only needs to expose its subtotal. At a split, the permitted left subtotal is the interval that satisfies the left bounds while leaving a feasible amount for the right:

The same construction with scalar summaries(7)
α=max⁡(Lℓ,D−Ur),β=min⁡(Uℓ,D−Lr),Dℓ=α+t(β−α),Dr=D−Dℓ,0<t<1.\begin{aligned}\alpha&=\max(L_\ell,D-U_r),\quad\beta=\min(U_\ell,D-L_r),\\[6pt]D_\ell&=\alpha+t(\beta-\alpha),\quad D_r=D-D_\ell,\quad 0<t<1.\end{aligned}
L and U are the sums of lower and upper bounds in each child. Recursing over these compatible subtotals preserves the total budget and every component bound.

The allocation study validates coordinates and volume factors on 312 instances with 2–40 variables. On 8,192 unseen outputs of a frozen predictor, the maximum total-budget error falls from 0.1915 to 7.11 × 10−15, without a component-bound violation. Euclidean projection has slightly lower prediction RMSE on those same outputs. Exact constraints and prediction accuracy remain distinct objectives. [Tables 29–30]

For programmable biology, RCC offers a way to separate learned design preferences from geometric requirements that every candidate must satisfy. A generator supplies proposals; a coordinate system enforces the specified constraints; an objective guides the search within them. Better biological objectives and experimental evidence are still needed to turn those valid structures into useful designs.

Sources and data

Read the paper

This article is based on the supplied 37-page manuscript, marked under review at ICLR 2027. The original overview and backbone comparison are reproduced without altering their content. Charts are redrawn from reported tables; the interactive bridge is an explanatory mathematical example. Prospective experiments, post-hoc controls, and development pilots are identified separately.

  1. Read the full manuscript ↗Recursive Constraint Compilation for Guided Protein Backbone Design. Anonymous supplied version; PDF, 37 pages.
  2. Download the chart data ↗Reported aggregate values from Tables 3, 20, and 23, with source locations and evaluation scope.
  3. Figure provenance ↗Source checksum, page numbers, and crop coordinates for the original figures. Equation numbers in this article are local to the article.