{"id":"c3dc2681-c8f9-458c-b5c7-f40bcc60a9ba","arxiv_id":"2507.04496","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":0.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of recent theory and applications of structural identifiability in compartmental models, including identifiable reparametrizations and graph-based criteria.","lead":"This paper reviews recent progress on structural identifiability of compartmental models, covering applications, reparametrization methods, and graph-theoretic criteria for linear models. It is a useful survey for modelers in epidemiology, oncology, and systems biology who need to know when parameters can be recovered from data.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: spot-checks of Theorem 5.1 and the reparametrization example confirm the survey's claims; no concrete error found.","rationale":"The paper is a survey whose central claim is faithful reporting of recent progress. The most prominent stated result, Theorem 5.1, survives direct small-case verification and parameter-count plausibility. The Example 2.2 reparametrization also checks out algebraically. The reader's weakest assumption, correctness of cited results, is indeed the only meaningful risk, but it is the standard risk for any survey and does not, in this case, rise to a load-bearing defect. The ACCEPT verdict with moderate confidence is appropriate; no change is needed.","tokens_in":9314,"tokens_out":25896,"duration_ms":250195,"concrete_test":"Independently derive the transfer function for the 3-compartment bidirected path with input at 1, output at 3, and no leaks; verify it equals a21*a23 / [s(s^2 + (a12+a21+a23+a32)s + (a21*a23 + a12*a32 + a21*a32))], confirming four parameters but only three observable coefficients and hence unidentifiability. If the computation instead yields a denominator with a nonzero constant term, the theorem's no-leak reasoning would need revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No load-bearing defect surfaced. The most important structural claim, Theorem 5.1, was tested against direct transfer-function computations for small bidirected trees. For input 1, output 3 on the bidirected path 1-2-3 with no leaks, the transfer function is a21*a23 / [s(s^2 + (a12+a21+a23+a32)s + (a21*a23 + a12*a32 + a21*a32))], giving four parameters but only three nonzero coefficients, so the model is unidentifiable, exactly as the theorem predicts. For distance 1 cases (e.g., input 1, output 2 on the same path), the numerator gains a factor (s + a32), restoring enough coefficients for generic local identifiability. The one-leak, distance-1 case similarly balances coefficient count. The directed-cycle leak-interlacing criterion is not re-derived, but no inconsistency was found. The Example 2.2 reparametrization was also checked algebraically: the companion-matrix coefficients (-k1*k4 + k3), -(k1*k2 + k4), and -(k1+k2) follow from P*A = C*P, so the displayed ODEs are correct. The only residual risk is the standard one for a survey: not every cited theorem was independently re-derived. That risk is inherent and does not undermine the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This survey reviews recent progress on the structural identifiability of compartmental models. It has three main strands: (i) applications in epidemiology, oncology, and other areas, with a summary of how practitioners adjust unidentifiable models; (ii) theoretical and algorithmic methods for reparametrizing unidentifiable models, illustrated by the worked Example 2.2; and (iii) graph-theoretic criteria for linear compartmental models, highlighted by Theorem 5.1, which states that a bidirected-tree model with one input and one output is generically locally identifiable if and only if it has at most one leak and the input-output distance is at most one, and that a directed-cycle model is generically locally identifiable if and only if its leaks are interlacing. The paper also collects recent classification results in Tables 1–5 and poses several open questions.","tokens_in":9550,"tokens_out":2988,"duration_ms":34835,"significance":"The paper is a well-structured survey that will be useful to both applied modelers and theorists. Its main strengths are the careful synthesis of disparate results into Tables 1–5, the explicit statement of Theorem 5.1 as a readily checkable structural criterion, and the worked reparametrization in Example 2.2, which illustrates the abstract material concretely. The survey does not claim new theorems, but it provides a clear map of what is known and what is open, and it correctly frames the scope of the results, noting, for example, that no complete answer exists even for linear compartmental models. The presentation is faithful to the cited literature as far as I could verify, and the central structural claim in Theorem 5.1 is supported by direct transfer-function computations for small examples. The paper is a valuable service to the community and is suitable for publication.","major_comments":[],"minor_comments":[{"comment":"The title in the arXiv header reads 'COMP AR TMENT AL MODELS' with extra spaces; this should be corrected to 'COMPARTMENTAL MODELS'.","section":"Title"},{"comment":"In the paragraph on ovarian follicle population dynamics, the Julia package is written as 'StucturalIdentifiability'; the correct spelling is 'StructuralIdentifiability'.","section":"Section 3"},{"comment":"In the sentence before Table 4, 'unidentifable' is missing an 'i' and should be 'unidentifiable'; similarly, Table 5's header 'parameters that unidentifiable' should read 'parameters that are unidentifiable'.","section":"Section 5"},{"comment":"The phrase 'distance from the input to output' is used without a formal definition in the text; since it is central to the theorem, a one-sentence definition or an explicit reference to the definition in [6] would improve accessibility.","section":"Theorem 5.1"}],"recommendation":"accept","confidential_remarks":"The paper contains numerous self-citations, but in this survey context they are appropriate because the authors are among the main contributors to the cited results. The only editorial concern is the title typo, which can be fixed at the proof stage. The survey fits well in a mathematical biology or applied mathematics venue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dan—This is a review paper, and it should be judged as one. No new theorems, no new algorithms, no data. What it does instead is pull together a decade of work on structural identifiability of compartmental models, organizes it by application and by theory, and gives the reader practical tables of identifiable/unidentifiable model classes. That is genuinely useful. The paper is well structured and the writing is clear. I spot-checked the central structural claim, Theorem 5.1 (bidirected-tree and directed-cycle criteria), against direct transfer-function computations for small cases, and it holds. The Example 2.2 reparametrization is also correct; the companion-matrix coefficients follow from P*A = C*P. So the stress-test note checks out. The survey's strongest value is in the tables (Tables 2-5) and the open questions, which give a researcher a quick map of what is known and where the gaps are.\n\nSoft spots are what you would expect from a survey. The classification in the tables rests on cited results, and the authors do not re-derive them. That is standard, but it means the survey's practical guidance is only as good as the cited proofs. The self-citations are numerous (roughly a third of the references are by the authors or close collaborators), but most are published, peer-reviewed results, and the ones I checked are correctly summarized. There is a typo in the title (COMP AR TMENT), which is cosmetic. A larger philosophical claim in the Outlook—that structural identifiability is 'in some sense a solved problem'—is an oversimplification; the paper itself then walks it back by noting that model uncertainty is the real issue. I'd take that line as a rhetorical opening rather than a substantive error.\n\nWho is this for? Practitioners in epidemiology, systems biology, and pharmacometrics who want to know whether their model is identifiable before running an estimation routine, and theorists looking for open problems. It deserves a serious referee. The verification burden is modest for a review, but an editor should still send it out to catch any mis-citations and to confirm the tables match the literature.\n\nRecommendation: send to peer review. Accept if the authors fix the title typo and possibly soften the 'solved problem' line.","headline":"A clean, faithful survey of structural identifiability for compartmental models; no new results but a genuinely useful synthesis that deserves review.","tokens_in":10048,"tokens_out":1975,"would_cite":true,"duration_ms":20367,"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":"This survey argues that structural identifiability of linear compartmental models is now predictable from the graph: for bidirected-tree models with one input and one output, identifiability holds exactly when there is at most one leak…","keywords":["structural identifiability","compartmental models","linear ODE models","directed graphs","leak-interlacing","identifiable functions","reparametrization","bidirected trees"],"falsifier":"Run a symbolic identifiability check on a small bidirected-tree model with one input, one output, and two leaks, such as a three-compartment tree with input at compartment 1, output at compartment 2, and leaks from compartments 1 and 3; if any such model is generically locally identifiable, Theorem 5.1 is false. A second test is to find a directed-cycle model whose leaks are not interlaced but whose parameters are all generically locally identifiable.","tokens_in":9088,"feed_emoji":"🧬","tokens_out":6506,"duration_ms":61402,"temperature":0.7,"pith_summary":"This survey argues that for linear compartmental models, structural identifiability—whether parameter values can be recovered, up to a finite set, from perfect input-output data—can often be read directly from the model's graph. It presents recent theorems and tables that give complete graph-theoretic classifications for several model classes, most prominently bidirected trees and directed cycles. It also reviews practical routes for dealing with unidentifiable models, including fixing parameters, adding outputs, simplifying the model, and reparametrizing using identifiable functions. The payoff it claims is that, for these classes, a researcher can look at the graph and know in advance whether generic parameter estimation is feasible.","feed_headline":"Graph shape alone decides identifiability of these model classes","feed_subtitle":"For bidirected trees and directed cycles, counting leaks and checking input-to-output distance settles whether parameters can be recovered.","key_machinery":"The central object is the directed graph attached to a linear compartmental model: vertices are compartments, directed edges are transfers between compartments, and designated node sets record inputs, outputs, and leaks (losses to the environment). The criteria that carry the argument are graph-theoretic conditions—counting leaks, measuring the distance from an input to an output, and checking the interleaving of leaks with inputs and outputs. For reparametrization, the working objects are identifiable functions of the parameters (monomials attached to paths and cycles in the graph) and the linear reparametrizations they generate, for which explicit formulas are available in single-in-single-out cases.","core_discovery":"The paper's central claim is that identifiability of linear compartmental models is determined by the structure of the underlying directed graph for a growing list of model classes. The flagship result, Theorem 5.1, states that a bidirected-tree model with one input and one output is generically locally identifiable if and only if it has at most one leak and the distance from the input to the output is at most one, and that a directed-cycle model is generically locally identifiable if and only if its leaks are interlaced—that is, between any two leaks there is an input or an output. The accompanying tables assert complete classifications of identifiable and unidentifiable classes, and identify whole classes of parameters that are globally identifiable or unidentifiable. If these classifications are correct, they turn a symbolic-computation question into a graph-inspection question for those model families.","pith_inferences":["If the graph-theoretic criteria hold, the practical design rule for linear compartment experiments is to put an output next to the input and avoid more than one leak, since those are exactly the conditions that guarantee generic local identifiability in bidirected trees.","The theorem naturally invites a multi-input/multi-output version for bidirected trees; such an extension is not proven in the paper, and testing it would require comparing symbolic identifiability results against the new graph conditions.","The same graph-first perspective could be applied to model selection: among many plausible compartment graphs, identifiability criteria could filter out models that cannot be calibrated before any data are collected.","For nonlinear compartmental models, no comparable structural classification is available, so the survey's graph-based approach is a specifically linear-model achievement rather than a general principle."],"forward_implications":["For any bidirected-tree model with one input and one output, identifiability can be decided by counting leaks and checking whether the input and output are adjacent or distance at most one.","For directed-cycle models, identifiability is equivalent to the leak-interlacing condition, with no dependence on the number of compartments.","The tables give computation-free identifiability verdicts for further classes such as catenary, mammillary, identifiable-path, and identifiable-cycle models, including at the level of individual parameters.","Unidentifiable linear compartmental models can often be repaired without changing the model by a linear reparametrization generated from identifiable functions, with explicit formulas in single-in-single-out cases.","Adding inputs or outputs is a reliable route to making a model identifiable, and the survey frames the open problem of finding minimal such configurations."],"supporting_citations":[{"why":"supplies the directed-cycle and catenary identifiability results that back Theorem 5.1 and Tables 2–3.","marker":"[1]"},{"why":"supplies the bidirected-tree identifiability criterion and the general input-output equation formula behind Theorem 5.1.","marker":"[6]"},{"why":"defines identifiable paths and cycles and provides several classifications of identifiable and unidentifiable path/cycle models in Tables 2, 3, and 5.","marker":"[7]"},{"why":"introduces identifiable paths/cycles and scaling reparametrizations that underpin the reparametrization section and identifiable-cycle definitions.","marker":"[38]"},{"why":"gives the algorithm and explicit formulas for identifiable linear reparametrizations, including the Example 2.2 demonstration.","marker":"[37]"},{"why":"provides the identifiability results for identifiable-cycle models and minimal input/output configurations used in Table 2 and the discussion.","marker":"[39]"},{"why":"characterizes unidentifiable parameters outside output-reachable subgraphs and operations preserving identifiability, used in Table 5.","marker":"[21]"},{"why":"supplies mammillary-model identifiability results and global identifiability of specific parameters in Table 4.","marker":"[13]"},{"why":"provides identifiability results for catenary systems, used for the global identifiability of all parameters in Table 4.","marker":"[14]"},{"why":"establishes conditions for the existence of scaling reparametrizations of linear compartment models cited in Section 4.","marker":"[2]"}],"fun_headline_variants":["Graph shape decides identifiability for tree and cycle compartmental models","For bidirected trees and cycles, graph shape alone settles identifiability","Identifiability: just count leaks and check distances for these models","Leak count and input-output distance predict identifiability in these models"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The survey's practical guidance assumes that the cited classification theorems are correct and that the listed model classes cover all relevant cases; in particular, Theorem 5.1 rests on earlier results about bidirected trees and directed cycles that the paper does not re-derive.","fun_headline_variants_meta":{"raw":{"variants":["Graph shape decides identifiability for tree and cycle compartmental models","For bidirected trees and cycles, graph shape alone settles identifiability","Identifiability: just count leaks and check distances for these models","Leak count and input-output distance predict identifiability in these models"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001861,"raw_usage":{"total_tokens":7233,"prompt_tokens":796,"completion_tokens":6437,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":412,"completion_tokens_details":{"reasoning_tokens":6360}},"tokens_in":412,"tokens_out":6437,"duration_ms":52837,"temperature":1.0,"reasoning_tokens":6360,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:45:00.628334+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a symbolic identifiability check on a small bidirected-tree model with one input, one output, and two leaks, such as a three-compartment tree with input at compartment 1, output at compartment 2, and leaks from compartments 1 and 3; if any such model is generically locally identifiable, Theorem 5.1 is false. A second test is to find a directed-cycle model whose leaks are not interlaced but whose parameters are all generically locally identifiable.","supporting_citations":[{"cited_title":"Identifiability of directed-cycle and catenary linear compartment models","cited_arxiv_id":"2412.05283","evidence_quote":"supplies the directed-cycle and catenary identifiability results that back Theorem 5.1 and Tables 2–3."},{"cited_title":"Identifiability of linear compartmental tree models and a general formula for input-output equations","cited_arxiv_id":null,"evidence_quote":"supplies the bidirected-tree identifiability criterion and the general input-output equation formula behind Theorem 5.1."},{"cited_title":"Identifiable paths and cycles in linear compartmental models","cited_arxiv_id":null,"evidence_quote":"defines identifiable paths and cycles and provides several classifications of identifiable and unidentifiable path/cycle models in Tables 2, 3, and 5."},{"cited_title":"Identifiable reparametrizations of linear compartment models","cited_arxiv_id":null,"evidence_quote":"introduces identifiable paths/cycles and scaling reparametrizations that underpin the reparametrization section and identifiable-cycle definitions."},{"cited_title":"Algorithm to find new identifiable reparametriza- tions of parametric rational ODE models","cited_arxiv_id":null,"evidence_quote":"gives the algorithm and explicit formulas for identifiable linear reparametrizations, including the Example 2.2 demonstration."},{"cited_title":"Identifiability results for several classes of linear com- partment models","cited_arxiv_id":null,"evidence_quote":"provides the identifiability results for identifiable-cycle models and minimal input/output configurations used in Table 2 and the discussion."},{"cited_title":"Linear compartmental models: input- output equations and operations that preserve identifiability","cited_arxiv_id":null,"evidence_quote":"characterizes unidentifiable parameters outside output-reachable subgraphs and operations preserving identifiability, used in Table 5."},{"cited_title":"Cobelli, A","cited_arxiv_id":null,"evidence_quote":"provides identifiability results for catenary systems, used for the global identifiability of all parameters in Table 4."},{"cited_title":"On the existence of identifiable reparametrizations for linear compartment models","cited_arxiv_id":null,"evidence_quote":"establishes conditions for the existence of scaling reparametrizations of linear compartment models cited in Section 4."}],"review_version":1}