{"id":"de0d69ff-93da-47fd-9084-a6bbc872d547","arxiv_id":"1908.03306","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A new algorithm blends SAXS spherical-average intensities with cryo-EM Fourier slices to reconstruct missing wedge information, demonstrated on simulated shape models.","lead":"Cryo-electron microscopy loses information when molecules sit in only a few orientations, leaving empty wedges in the 3D picture. This paper blends it with SAXS data, a low-resolution all-angle scattering signal, to patch those wedges on simulated shapes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 3D fusion step is underdetermined: SAXS contributes a single scalar per shell, while the Moore-Penrose solution arbitrarily sets all null-space coefficients of the EM sampling operator to zero, so the missing-wedge values are not actually determined by SAXS.","rationale":"I agree with the reader's conditional verdict, but I locate the most load-bearing weakness elsewhere. The reader emphasizes known orientations and noise, which are real practical limitations; however, the paper is a noiseless proof-of-concept and explicitly treats orientations as given, so those issues are addressable by adding experiments. The more fundamental structural problem is that the 3D algorithm's use of the Moore-Penrose pseudo-inverse collapses the underdetermined EM sampling problem to a single one-parameter family, so the SAXS intensity, which is a scalar per shell, cannot actually determine the many angular degrees of freedom in the missing wedge. The paper itself flags the great-circle sampling difficulty and the minimum-norm nature of the pseudo-inverse, but it never proves that the SAXS-sphere intersection is two points rather than a continuum. If the null-space dimension is positive for the demonstrated sampling patterns, the claimed success in §5.3 and §5.4 is not evidence that SAXS fills blind spots; it is evidence that a particular smooth interpolation with a norm-scaled zeroth coefficient can look better than a different interpolation. This reinforces the reader's conditional verdict: the symmetry-relation typo and the missing null-space analysis should both be corrected before the central 3D claim is accepted. I set verdict_should_be to UNCHANGED because the reader already prescribed the correct level of scrutiny; my concern provides additional, more structural support for that conditionality rather than a reason to reject the entire paper outright.","tokens_in":21361,"tokens_out":9745,"duration_ms":112789,"concrete_test":"For the 3D smiley or minion sampling geometry in §5.3/§5.4, pick one Fourier shell and form the EM sampling matrix B from the stated great-circle points. Compute rank(B) and the dimension of the null space of the affine system (24). Then generate two or more coefficient vectors that differ only in the null-space components, adjust ρ̃00 for each via (26) so that both satisfy the SAXS intensity I(r), and compare their Fourier values inside the missing wedge. If these values differ materially, the SAXS constraint does not determine the blind spots and the reported reconstruction is an arbitrary minimum-norm choice rather than a SAXS-driven fusion result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central 3D claim requires that SAXS intensity I(r) together with EM samples on great circles determines the Fourier coefficients in the missing wedge. In §3.3.2 the authors reduce the EM fit to the linear system (24) and solve it with the pseudo-inverse F†. But when EM samples lie on great circles, the sampling operator B is rank-deficient even when the number of samples reaches N², as the paper itself notes by citing Freeden et al. Consequently, the affine solution set of (24) has a nontrivial null space: the general solution is F†(ρ̂ − ρ̃00/√(4π) 1) plus any vector in the null space of F. The SAXS constraint (26) is then only one quadratic equation in ρ̃00; it does not constrain the null-space components. The algorithm silently chooses these components to be zero (the minimum-norm solution), so the reconstructed values inside the missing wedge are artifacts of that choice, not information extracted from SAXS. Section 4.4 acknowledges that the pseudo-inverse selects the minimum-norm solution, but it does not justify that choice physically or show that the intersection of the EM affine subspace with the SAXS sphere is just two points, which is what the two-root selection procedure assumes. In 2D the counting works out (2n EM values plus one SAXS value for 2n+1 coefficients), but the 3D generalization, which is the claimed practical advance, does not. Thus, as written, the method does not establish that SAXS fills blind spots in 3D; it establishes that one particular least-norm interpolation, with ρ̃00 adjusted to match the SAXS intensity, can improve a toy reconstruction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a method for fusing SAXS and cryo-EM data in Fourier space. For each radius shell, EM projection data are treated as samples of the Fourier transform on lines (2D) or great circles (3D), while SAXS provides the spherical average of the squared Fourier modulus on that shell. The method expands the Fourier transform in Fourier series or spherical harmonics, combines the linear EM equations with the quadratic SAXS intensity constraint, and solves for the zeroth-order coefficient through a quadratic equation. Numerical experiments on Gaussian-based 2D and 3D toy models report improved reconstruction over EM alone when projection directions are missing. The paper also derives an intensity threshold I* below which SAXS is claimed to add no information to EM.","tokens_in":21675,"tokens_out":10618,"duration_ms":113779,"significance":"If correct, the approach would be a useful contribution to hybrid structural biology, and the lower-bound result (IEM = I*) is a clean and potentially interesting observation. The paper presents explicit formulas and fairly detailed numerical protocols for the toy models. However, the mathematical foundation has two serious gaps: an incorrect Hermitian-symmetry relation in both the 2D and 3D derivations, and an unaddressed null-space ambiguity in the 3D reconstruction. Because the central claim that SAXS fills missing wedges in 3D rests on these steps, the paper in its current form does not establish its main result.","major_comments":[{"comment":"The symmetry relation stated in Eq. (2), ρ̂(θ) = ρ̂(θ+π), is incorrect for a real-valued density. The correct identity is ρ̂(-ω) = ̅ρ̂(ω), so on the unit circle one has ρ̂(θ+π) = ̅ρ̂(θ). Consequently Eq. (6) should read ρ̂_{-k} = (-1)^k ̅ρ̂_k rather than ρ̂_{-k} = (-1)^k ρ̂_k, and the 3D analogues (20)-(21) require the same complex conjugation. The construction of the vectors and the derivation of the quadratic equations (12) and (26) use the incorrect relation, so the counting of independent real unknowns and the claimed closure of the system by the SAXS constraint are not valid for a general real-valued density. This is a load-bearing error because the entire fusion procedure is built on these symmetry reductions.","section":"Sec. 3.1, Eq. (2), Eq. (6) and Sec. 3.3.2, Eqs. (20)-(21)"},{"comment":"The 3D system is underdetermined in a way that the SAXS constraint does not close. The paper itself notes, citing Freeden et al., that N^2 point values on a great circle are not sufficient to determine a degree-(N-1) spherical polynomial, so the matrix F in (24) has a nontrivial null space. The Moore-Penrose solution (25) selects one particular affine solution, namely the minimum-norm one; the general solution is F†(ρ̂ - ρ̂_{0,0}1/√(4π)) + z with z ∈ null(F). The SAXS intensity is a single quadratic equation in ρ̂_{0,0} through (22)/(26) and imposes no condition on z. Thus the intersection of the EM affine subspace with the SAXS sphere is generically a positive-dimensional sphere, not two points as assumed in Section 4.2 and in the root-selection procedure. Section 4.4 explicitly acknowledges the minimum-norm choice but offers no physical justification for it. The missing-wedge coefficients are therefore set by the pseudo-inverse, not determined by the SAXS data, and the central 3D claim in the abstract and in Section 3.3.2 is not established.","section":"Sec. 3.3.2, Eqs. (24)-(26) and Sec. 4.2/4.4"},{"comment":"The root-selection procedure is heuristic and is not demonstrated to be reliable in the settings the paper targets. The selector in Eqs. (40)-(41) approximates ρ̂_{0,0} by directly summing the EM samples; when the missing wedge is large or the samples are clustered, this sum is biased and there is no guarantee that the selected root is the true one. The numerical experiments assess success against the known analytic ground truth, the experiments are noiseless, and the same model is used for both data generation and evaluation, but the method itself has no independent criterion to certify that the chosen root and the null-space components are correct. This is a practical limitation that should be acknowledged explicitly and, if possible, addressed.","section":"Sec. 4.5 and Sec. 5 numerical protocols"}],"minor_comments":[{"comment":"The expression for the Fourier transform of a Gaussian appears to be missing the dot product in the phase term: it should read exp(-ω^T Σω/2 - i μ·ω) or exp(-ω^T Σω/2 - i μ^Tω).","section":"Appendix A, Eq. (46)"},{"comment":"The text says 'l ∈ {1,2,3,...}' when introducing the truncated spherical harmonic expansion, but the sum includes l=0 through l=N-1, including ρ̂_{0,0}; the range should be stated consistently.","section":"Sec. 3.3.2"},{"comment":"The definition of 'bad points' as points with error greater than 20% of the original value is arbitrary and the threshold is not justified; the reported numbers should be accompanied by a precise error metric and, ideally, quantitative error bars or a table.","section":"Sec. 5.3.1"},{"comment":"The manuscript contains numerous typos and grammatical errors (e.g., 'pseduo inverse', 'Legenrde', 'to to select', 'co-origin'), which should be corrected in a revision.","section":"Throughout"}],"recommendation":"reject","confidential_remarks":"I recommend rejection. The two mathematical issues in my major comments are fundamental: the incorrect Hermitian-symmetry relation affects the 2D illustrative derivation and the 3D generalization, and the null-space ambiguity means the 3D method as described does not actually determine the missing wedge from SAXS data. The numerical experiments are suggestive but cannot compensate for the fact that the central claim is not supported by the presented mathematics. A revision would require a substantially new algorithmic idea for the 3D case, not merely a correction of local errors."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know that this paper has a genuinely new idea and a genuine flaw. The new idea: model the Fourier transform on each shell as a truncated harmonic expansion, use EM samples on great circles to fix most coefficients, then use the per-shell SAXS intensity to solve for the remaining DC term. The lower-bound result in Section 4.4 is a nice piece of analysis: SAXS only helps when I(r) > I*(r), where I* is the minimum norm compatible with EM samples. That is a real contribution.\n\nThe flaw is more serious than the paper lets on. In 3D, the EM sampling operator on great circles is rank deficient, so the linear system has a large null space. SAXS gives one scalar per shell—the squared norm of the coefficient vector. That constraint does not pick out the missing-wedge values; it only fixes the total energy. The algorithm uses the Moore-Penrose pseudo-inverse, which silently sets all null-space components to zero. The two-root solution is an artifact of writing the quadratic in the single DC coefficient; the rest of the null sphere remains arbitrary. So the claim that SAXS fills blind spots in 3D is not established. What the method actually does is compute the minimum-norm interpolation matching the EM samples, then adjust the DC term to match the SAXS intensity.\n\nThere is also a wrong symmetry relation early on: Eq (2) says \\hatρ(θ)= \\hatρ(θ+π), and Eq (20) says \\hatρ(u)=\\hatρ(-u). For a real density the correct relations involve complex conjugation: \\hatρ(-ω)=\\hatρ(ω)*. The derived coefficient relations (6) and (21) are off by that conjugate. This does not necessarily kill the numerics, but it reveals a misunderstanding of Fourier symmetry and needs to be fixed.\n\nThe numerical experiments are all noiseless toy models with known orientations, so they don't address the real-world bottlenecks of cryo-EM. The improvements shown are real but likely due to the min-norm prior plus the intensity adjustment, not to SAXS actually determining Fourier coefficients in the missing wedge.\n\nWho should read this: people working on hybrid SAXS-EM approaches. The I* result is worth keeping. But the central 3D claim needs a rethink. I'd send it to peer review, because the idea is relevant and the flaw is subtle enough that a good referee conversation could sharpen it—maybe into a more modest claim about regularized interpolation with a SAXS norm prior.\n\nFor your own work: I wouldn't cite the fusion claim as-is, but I'd keep the lower-bound result in mind.","headline":"Clever per-shell SAXS norm constraint, but the 3D missing-wedge claim doesn't hold up: SAXS gives one scalar per shell while the EM null space stays arbitrary, and the paper's Fourier symmetry relations are wrong.","tokens_in":22265,"tokens_out":5016,"would_cite":false,"duration_ms":53866,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"SAXS curves fill blind spots in cryo-EM maps by anchoring Fourier space with a norm constraint.","keywords":["SAXS","cryo-EM","data fusion","Fourier slice theorem","spherical harmonics","missing wedge","single-particle reconstruction","structure determination"],"falsifier":"Take a synthetic density with a known Fourier transform, sample EM great circles with a deliberate angular gap, and compute the SAXS curve $I(r)$ exactly; then run the fusion and compare the filled Fourier shell with the true values. The claim predicts the interpolation error in the gap drops toward zero as the number of great circles grows, and that fusion always beats EM-alone whenever $I(r) > I^*(r)$ — a direct numerical check. Conversely, on an experimental data set with a known missing wedge, if adding real SAXS data does not reduce the reconstruction error inside the wedge relative to EM alone, the central claim would be refuted.","tokens_in":21144,"feed_emoji":"🔬","tokens_out":7067,"duration_ms":63628,"temperature":0.7,"pith_summary":"The paper argues that SAXS scattering curves, despite being low-resolution spherical averages, carry enough Fourier information to fill wedge-shaped gaps left by cryo-EM when the range of particle projection directions is limited. In the paper's picture, both experiments sample the same three-dimensional Fourier transform of the electron density: EM provides values on planes through the origin (great circles on each sphere in Fourier space), while SAXS provides, on each sphere, the integral of the squared magnitude. The authors show these two partial views can be combined, sphere by sphere, by writing the Fourier transform as a truncated spherical-harmonic expansion and solving a quadratic equation that enforces the SAXS norm. If the method works as claimed, cryo-EM reconstructions with anisotropic coverage, a known failure mode in single-particle analysis, can be repaired with a relatively cheap additional SAXS measurement. The authors demonstrate the effect in numerical experiments on 2D and 3D synthetic shapes, including a model of a 'minion' molecule.","feed_headline":"SAXS curves fill blind spots in cryo-EM maps","feed_subtitle":"The paper shows SAXS intensity completes the Fourier shells that cryo-EM leaves empty.","key_machinery":"The load-bearing object is the truncated spherical-harmonic expansion of the Fourier transform restricted to each concentric sphere, together with the SAXS intensity written as a squared-norm constraint. On a sphere of radius $r$, EM projection data give the values of $\\hat{\\rho}$ at points lying on great circles; these become a linear system $B\\tilde{\\rho} = \\hat{\\rho}$ for the expansion coefficients. The SAXS measurement $I(r) = \\int_{S^2} |\\hat{\\rho}(ru)|^2 du$ becomes the quadratic constraint $\\|\\tilde{\\rho}\\|^2 = I(r)$ (up to normalization), which turns the underdetermined system into a solvable one: eliminating all coefficients except the real-valued zeroth one leaves a quadratic equation, and the remaining coefficients follow linearly. Two devices carry the argument: the relation between the two roots (they differ by a unitary transformation whose difference lies in the null space of the sampling matrix, so no extra data can distinguish them without a selection rule) and the identity $I^* = I_{\\mathrm{EM}}$, obtained via the Sherman-Morrison formula, which sets the threshold below which the SAXS constraint is redundant or inconsistent.","core_discovery":"The central claim is that SAXS data, which record only the spherical average of the squared electron-density Fourier magnitude, can be used to fill in missing wedge regions of Fourier space in cryo-EM reconstructions when projection directions are limited. Concretely, the paper derives a per-sphere reconstruction scheme: on a sphere of radius $r$ in Fourier space, the EM samples fix values of the transform on several great circles; the SAXS curve fixes the integral of $|\\hat{\\rho}|^2$ over the whole sphere. Expanding the restricted transform in spherical harmonics up to order $N-1$, the EM samples give an underdetermined linear system in the expansion coefficients, and the SAXS intensity supplies the missing constraint, closing the system through a quadratic equation for the zeroth coefficient. After choosing the correct root, all coefficients are determined, the transform is interpolated on that sphere, and the full density is recovered by inverse Fourier transform. The paper also shows that for each radius there is a minimal SAXS intensity $I^*$ below which the quadratic equation has no real root and the fusion degenerates to a plain EM reconstruction, so SAXS only adds information when its measured intensity exceeds this per-shell threshold.","pith_inferences":["The same per-sphere quadratic formulation could apply to other modalities that provide only angular averages, such as neutron scattering or light scattering, whenever matching Fourier slices are available.","The fusion gain should depend on how much Fourier amplitude sits inside the missing wedge; the paper's angle-dependent gap experiments suggest a quantitative version of this statement could be derived.","A natural stress test is to inject realistic EM noise and orientation error into the simulations; graceful degradation of the root selection would indicate a path to experimental use."],"forward_implications":["A single SAXS measurement could be combined with an existing cryo-EM reconstruction to fill missing projection wedges, removing a known source of anisotropy artifacts in single-particle analysis.","The per-radius threshold $I^*$ gives a practical diagnostic: before running the fusion, compute the EM intensity from the samples and check that the measured SAXS intensity exceeds it; only shells where it does can benefit from fusion.","The root-selection rule based on stability under resampling is reusable in any underdetermined-plus-norm reconstruction and, on the paper's numerical evidence, becomes reliable as the EM sampling is refined.","The 3D version of the procedure has the same structure, so the benefit should carry over to macromolecular complexes as long as the orientation assignment from the prior EM reconstruction is accurate."],"supporting_citations":[{"why":"Supplies the mathematical model of SAXS data acquisition as spherical averaging of the Fourier magnitude.","marker":"Dong et al. [2015a]"},{"why":"Frames SAXS and cryo-EM as cross-validating partial views of the same Fourier spectrum, the starting point of the fusion.","marker":"Afsari et al. [2015]"},{"why":"Establishes the compatibility conditions under which SAXS and cryo-EM data can be combined.","marker":"Kim et al. [2017]"},{"why":"Provides the matrix-inverse lemma used to prove that the EM-only intensity equals the threshold $I^*$ below which fusion degenerates.","marker":"Sherman and Morrison [1950]"},{"why":"Supplies the constructive-approximation result that points on one great circle cannot determine a spherical polynomial, motivating the SAXS constraint.","marker":"Freeden et al. [1998]"},{"why":"Offers the fast spherical Fourier transform used as an alternative solver in the 3D procedure.","marker":"Kunis and Potts [2003]"},{"why":"Offers the non-equispaced FFT used to solve the 2D linear system and interpolate.","marker":"Potts et al. [2001]"}],"fun_headline_variants":["SAXS fills missing Fourier shells in cryo-EM","Cryo-EM's blind spots solved by SAXS intensity","SAXS per-shell fusion fills cryo-EM gaps","SAXS data rescues cryo-EM from missing views","SAXS closes the wedge cryo-EM misses"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole construction presumes the particle orientations are already fixed by a prior EM reconstruction and that the EM Fourier samples and SAXS intensity come from the same noiseless molecular conformation — if the orientations are wrong or the data are noisy, the per-sphere quadratic equation and the choice between its two roots lose their justification.","fun_headline_variants_meta":{"raw":{"variants":["SAXS fills missing Fourier shells in cryo-EM","Cryo-EM's blind spots solved by SAXS intensity","SAXS per-shell fusion fills cryo-EM gaps","SAXS data rescues cryo-EM from missing views","SAXS closes the wedge cryo-EM misses"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000513,"raw_usage":{"total_tokens":2504,"prompt_tokens":968,"completion_tokens":1536,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":1453}},"tokens_in":584,"tokens_out":1536,"duration_ms":11841,"temperature":1.0,"reasoning_tokens":1453,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:18:37.955282+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a synthetic density with a known Fourier transform, sample EM great circles with a deliberate angular gap, and compute the SAXS curve $I(r)$ exactly; then run the fusion and compare the filled Fourier shell with the true values. The claim predicts the interpolation error in the gap drops toward zero as the number of great circles grows, and that fusion always beats EM-alone whenever $I(r) > I^*(r)$ — a direct numerical check. Conversely, on an experimental data set with a known missing wedge, if adding real SAXS data does not reduce the reconstruction error inside the wedge relative to EM alone, the central claim would be refuted.","supporting_citations":[{"cited_title":", and Chirikjian, G.\\,S","cited_arxiv_id":null,"evidence_quote":"Frames SAXS and cryo-EM as cross-validating partial views of the same Fourier spectrum, the starting point of the fusion."},{"cited_title":"Cross-validation of data compatibility between small angle x-ray scattering and cryo-electron microscopy","cited_arxiv_id":null,"evidence_quote":"Establishes the compatibility conditions under which SAXS and cryo-EM data can be combined."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the matrix-inverse lemma used to prove that the EM-only intensity equals the threshold $I^*$ below which fusion degenerates."},{"cited_title":"Constructive Approximation on the Sphere","cited_arxiv_id":null,"evidence_quote":"Supplies the constructive-approximation result that points on one great circle cannot determine a spherical polynomial, motivating the SAXS constraint."},{"cited_title":"Fast spherical Fourier algorithms","cited_arxiv_id":null,"evidence_quote":"Offers the fast spherical Fourier transform used as an alternative solver in the 3D procedure."},{"cited_title":"Fast Fourier transforms for nonequispaced data: a tutorial","cited_arxiv_id":null,"evidence_quote":"Offers the non-equispaced FFT used to solve the 2D linear system and interpolate."}],"review_version":1}