A custom AI reviewer for translational dose relevance

PerfectPaper’s translational dose agent reads a preclinical therapeutics manuscript for the objection that decides whether a clinical development argument lands: whether the concentrations and doses that produced the effect are achievable in a patient. The agent checks in vitro concentrations against reported human exposure, animal dosing against a stated conversion basis, and whether target engagement was ever demonstrated at the effective dose.

Copy the brief below into a custom agent slot.

The classic form of this objection is short and hard to answer: the effect requires ten micromolar, and the achievable plasma concentration is two hundred nanomolar. Nothing in the experiment is wrong. The dose-response curve is clean, the replicates agree, the statistics are correct, and the number on the x-axis is fifty times what a patient will ever see. This is the reason a translational objection cannot be answered with a re-analysis: the comparison a reviewer is making uses a number that is not in the manuscript, and adding animals, replicates or significance to the number that is in the manuscript changes nothing about the comparison.

The literature carries several worked demonstrations of exactly this failure at scale. He and Wondisford, writing on metformin in Cell Metabolism 2015;21:159–162, state that “almost all previous studies were conducted with supra-pharmacological concentrations (doses) of metformin, 10-100 times higher than maximally achievable therapeutic concentrations found in patients with type 2 diabetes mellitus” — a whole mechanistic literature built above the achievable range. Nelson and colleagues reached the same conclusion about curcumin in Journal of Medicinal Chemistry 2017;60:1620–1637, classifying it as both a pan-assay interference compound and an invalid metabolic panacea, noting that its apparent in vitro and in vivo activity had produced more than 120 clinical trials, and reporting that no double-blind, placebo-controlled trial of curcumin has succeeded. The agent exists to make that comparison a step in drafting rather than a finding in a decision letter.

When to use this agent

• Your manuscript argues that a compound merits clinical development
• Effects were demonstrated in vitro at concentrations you have not compared to human exposure
• Animal doses were chosen from prior literature rather than from exposure matching
• The compound is a repurposed drug with known human pharmacokinetics
• Your discussion contains the phrase clinically relevant
• No plasma, tissue or tumour concentrations were measured in the in vivo experiments
• The in vitro work was performed in medium containing 10% fetal bovine serum and the comparator is a total plasma concentration
• A dose in mg/kg appears in the abstract with no human equivalent
• The compound is a natural product, a polyphenol or a nutraceutical, where high-micromolar activity and low-nanomolar exposure are the common pattern
• A reviewer has already written that the model is not appropriate or that the effect size is not meaningful

What translational dose relevance means

Translational dose relevance is the question of whether the exposure that produced an effect in a dish or an animal is an exposure a person could be given, and it is a separate question from whether the effect is real. A compound can inhibit its target with a genuine, reproducible, mechanistically explained IC50 of 10 micromolar and still be undevelopable at that target, because 10 micromolar of free drug at the tissue is not reachable at a tolerated dose. The finding survives; the development argument does not.

The governing principle is the free drug principle: the pharmacological effect at a target is set by the concentration of unbound drug in the aqueous phase surrounding that target, not by the total measured concentration and not by the bound fraction. Smith, Di and Kerns set out the practical consequences in Nature Reviews Drug Discovery 2010;9:929–939, and their central correction is one manuscripts still get wrong: “in vivo efficacy is determined by the free (unbound) drug concentration surrounding the therapeutic target, not by the free drug fraction.” Two numbers therefore have to be made comparable before any ratio means anything — the free concentration that produced the effect, and the free concentration achievable in a patient.

The free drug principle assumes the compound crosses the membrane between blood and target by passive diffusion until the two unbound concentrations equilibrate, and that assumption is the part that fails for a substrate of an uptake transporter. Metformin is the standard illustration, and it is the same compound the objection above is built on: hepatocytes take it up across the sinusoidal membrane through the organic cation transporter OCT1, so the liver sees more drug than a plasma sample reports. The size of that enrichment is measurable rather than rhetorical. Morse and colleagues, dosing cynomolgus monkeys in The AAPS Journal 2017;19:1878–1889, reported an in vivo unbound liver-to-plasma partition coefficient — Kp,uu — of 15 for metformin, against the value of 1 that passive equilibration predicts. Sogame and colleagues found the same direction in rats in Biomedicine & Pharmacotherapy 2011;65:451–455, where a 50 mg/kg oral dose gave liver-to-plasma concentration ratios of 4.2 for metformin and 61.3 for phenformin.

Note the size of that exception, though, because a discussion section invoking it usually implies a larger one. Roughly an order of magnitude of hepatic enrichment narrows a ten-to-hundred-fold shortfall; it does not erase one, and it says nothing whatever about a target outside the liver. The free-versus-free comparison this agent runs is therefore a floor on target-site exposure for a passively distributed drug, not necessarily its ceiling for a transporter substrate — and a manuscript relying on the exception has to name the transporter, name the tissue and give the measured partition coefficient, rather than invoking transport in general as a reason the plasma number can be set aside.

This is a construct question about the argument, not an execution question about the experiment, which is why it behaves like the model relevance objection and unlike a statistics objection. A tighter confidence interval around an IC50 of 10 micromolar makes the translational problem worse, by lending precision to a claim that exposure cannot support. The same asymmetry makes a mechanism-not-shown comment expensive: both are objections to what the design can demonstrate rather than to how well it was executed.

Why the comparison is almost always made against the wrong number

The comparison in most manuscripts is a free in vitro concentration set against a total plasma concentration, and that mismatch overstates achievable exposure by exactly the factor by which the drug is bound. A compound that is 99% bound in human plasma has a fraction unbound of 0.01, so a reported total Cmax of 10 micromolar corresponds to 100 nanomolar of free drug. Setting a 1 micromolar in vitro IC50 against the total figure makes the drug look ten-fold better than achievable; setting it against the free figure shows it is ten-fold short. The direction of the error is always the same, and it is always in the author’s favour, which is why reviewers check it first.

Two mechanisms make the mismatch systematic rather than occasional. First, standard culture medium contains roughly 10% fetal bovine serum, so the protein content of the assay is about a tenth of plasma and the fraction unbound in the well is far higher than the fraction unbound in a patient — an assay IC50 is therefore close to a free concentration by construction. Second, human pharmacokinetic parameters are published as total concentrations, because that is what a validated bioanalytical assay measures. The author reads a free number from their own figure, reads a total number from a package insert or a pharmacokinetic paper, and divides.

The correction requires the fraction unbound in human plasma, measured by rapid equilibrium dialysis or ultrafiltration and reported in the compound’s pharmacokinetic literature. Where that value is published, the agent asks for the corrected comparison. Where it is not, the agent says so rather than estimating it, because a fabricated fu is a worse defect than a missing one. There is a second, less common form of the same error running the other way: an assay performed in medium supplemented to physiological protein levels, or in whole blood, produces an IC50 that is already protein-adjusted, and comparing that to a free plasma concentration double-counts the binding.

Two adjacent details belong in the same check. Metabolites frequently exceed parent drug in circulation and do not appear in the in vitro experiment at all — Boocock and colleagues reported in Cancer Epidemiology, Biomarkers & Prevention 2007;16:1246–1252 that the area under the plasma concentration curve for resveratrol-3-sulfate and the resveratrol monoglucuronides was “up to 23 times greater” than that of resveratrol itself. And the compartment that matters is the one containing the target: a tumour, a brain, an intracellular kinase and a plasma sample are four different concentrations, and only one of them is routinely reported.

Common questions about protein binding and achievable concentration

What concentration actually counts as clinically achievable?

One at or below the free, unbound plasma concentration reached at a tolerated human dose — not the total plasma concentration printed in a pharmacokinetic table. Most drugs are substantially protein bound, so the comparison has to be free-to-free or total-to-total; setting a free in vitro concentration against a total plasma value overstates achievable exposure, always in the same direction.

Do I need to correct for protein binding when I compare my IC50 to a plasma concentration?

Yes, whenever the in vitro number is close to a free concentration and the human number is a total plasma concentration from a standard bioanalytical assay, which is the usual case. Multiply the total plasma concentration by the fraction unbound to get the free concentration the target actually sees; skipping this step is the single most common reason a translational comparison looks better than it is.

Why is my compound’s IC50 so much lower in serum-free medium than in medium with fetal bovine serum added?

Because fetal bovine serum supplies most of the protein in the well, and standard culture medium contains roughly 10% serum — about a tenth of the protein content of human plasma. A compound that binds protein shows a more potent, lower IC50 with less serum present; the serum-free number sits closer to a free concentration, and the FBS-containing number sits closer to what a total-plasma comparison would show.

The ratio, and what each order of magnitude does to the claim

Roughly an order of magnitude is the threshold at which the ratio between effective concentration and achievable free exposure stops reading as a caveat and becomes the central limitation of a translational argument, and PerfectPaper’s dose agent reports that ratio in those terms rather than leaving it implicit. State the ratio as a single number, because that number, not the qualitative claim, is what a reviewer computes.

The tenfold threshold is a heuristic, not a universal cutoff, and a specialist reviewer applies it differently depending on the shape of the concentration-response relationship. A steep Hill slope — the kind seen at an allosteric site or a cooperative multimeric target — means a modest shortfall in free concentration can already collapse occupancy from near-saturating to negligible, so the objection can bite well before an order of magnitude. A target with spare capacity, where a fraction of maximal occupancy already produces the full pharmacological effect, tolerates a larger shortfall than the heuristic implies. State which regime applies, because “roughly tenfold” is a starting point for the reviewer’s arithmetic, not a substitute for it.

The published cases below are useful calibration precisely because each was argued in print by people who believed the translational claim.

The ivermectin case is the cleanest available demonstration of what the comparison is for, because both halves were published within months of each other. The in vitro paper reported an approximately 5,000-fold reduction in viral RNA at 48 hours in Vero-hSLAM cells and recommended further investigation. The pharmacokinetic paper then set the same inhibitory concentration against simulated human plasma profiles and concluded that “the likelihood of a successful clinical trial using the approved dose of ivermectin is low,” adding that combination therapy should be evaluated in vitro instead. Neither paper contains an error. The second paper contains the comparison the first one omitted, and the comparison is what decided the question.

A resveratrol worked example makes the arithmetic concrete. A single 5 g oral dose — a dose far above anything a chronic regimen would use — produced a mean peak plasma resveratrol of 2.4 micromol/L, and that is a total concentration measured at one time point 1.5 hours after dosing. An in vitro experiment run at 50 micromol/L therefore sits roughly twenty-fold above the highest total plasma concentration a 5 g dose has been shown to produce, before any correction for protein binding, for the fall from Cmax over the dosing interval, or for the 77% of urinary agent-derived species Boocock and colleagues recovered within 4 hours at the lowest dose. Report that arithmetic yourself and the discussion is a limitation. Omit it and the discussion is an overclaim, which is the same defect an overclaim reviewer catches from the language side.

Common questions about the ratio and whether it’s fatal

How far above achievable exposure does a concentration stop being clinically achievable?

Roughly an order of magnitude — tenfold — is where the ratio between effective concentration and achievable free exposure stops reading as a caveat and becomes the central limitation of the translational argument. Below that threshold the gap is usually defensible as assay variability or a conservative exposure estimate; above it, a reviewer will name the ratio explicitly, and so should the manuscript.

Is a concentration that isn’t clinically achievable a fatal flaw for my paper?

No. A high effective concentration limits the translational claim, not the mechanistic finding, and the two are separable in an afternoon of editing. Reporting the ratio and reframing the paper as mechanistic rather than translational — “the present work characterises the mechanism rather than proposing the compound as a clinical candidate” — is the remedy that survives review.

Where the animal dose came from, and why “precedent” is not an answer

An in vivo dose has one of four defensible origins — exposure matching to a known human exposure, body surface area or allometric conversion, the maximum tolerated dose in that species, or a stated pharmacodynamic target — and a fifth, undefensible one: the dose used in the paper the authors read first. The agent reports doses chosen by precedent alone, because a dose inherited through a citation chain carries no relationship to human exposure at any link in the chain.

The standard conversion is arithmetic a reviewer performs in ten seconds. The FDA guidance Estimating the Maximum Safe Starting Dose in Initial Clinical Trials for Therapeutics in Adult Healthy Volunteers (Center for Drug Evaluation and Research, July 2005) gives body surface area normalisation factors in Table 1: the mouse km is 3 and the human km is 37, so a mouse dose in mg/kg is divided by 12.3 to give the human equivalent dose in mg/kg, against a reference human body weight of 60 kg. A tumour experiment at 100 mg/kg daily in mice implies a human equivalent dose near 8 mg/kg per day, about 490 mg daily for a 60 kg adult. The divisor is the species km divided into 37, so the same table gives 6.2 for rat, 1.8 for dog and 3.1 for monkey — a distinct number from the km itself, which is 6, 20 and 12 for those three species. Quoting the km where the divisor belongs is a transcription error the agent looks for, and in the dog it is the difference between 1.8 and 20.

Then state the limitation that manuscripts using this factor almost never state. That guidance is a safety document for setting a maximum recommended starting dose from a no-observed-adverse-effect level, and it says of its own scope that it “does not address dose escalation or maximum allowable doses in clinical trials.” It also directs that exposure measurements are not the basis for a first-in-human starting dose: “Measurements of systemic levels or exposure (i.e., AUC or Cmax) cannot be employed for setting a safe starting dose in humans.” A human equivalent dose computed this way is a starting-dose safety calculation with a further safety factor of 10 conventionally applied to it — it is not evidence that the efficacious exposure is achievable. Using it as though it were is a category error the agent names explicitly, and the guidance’s own reasoning says why: it retains the body surface area exponent b = 0.67 over the b = 0.75 that fits maximum tolerated doses better, on the stated grounds that “mg/m2 normalization provides a more conservative conversion.” Conservative for safety means systematically low for efficacy.

Biologics take a different route entirely, and a manuscript that applies the mouse km factor to an antibody has used the wrong instrument. The EMA Guideline on strategies to identify and mitigate risks for first-in-human and early clinical trials with investigational medicinal products (EMEA/CHMP/SWP/28367/07 Rev. 1, 2017) is the reference for the minimal anticipated biological effect level approach, which derives a starting dose from receptor occupancy and target-binding pharmacology rather than from body surface area. Route compounds the problem in both classes: intraperitoneal dosing has no clinical counterpart for most small molecules, and a manuscript that dosed intraperitoneally and discusses oral therapy has skipped a conversion it never named. Formulation carries the same weight, since a compound delivered in DMSO, in a cyclodextrin vehicle or as a nanoparticle is not the compound a patient swallows.

Questions about mouse-to-human dose conversion

How do I convert a mouse dose in mg/kg to a human dose?

Divide the mouse dose in mg/kg by 12.3, using the FDA’s July 2005 body-surface-area method (Table 1, mouse km = 3, human km = 37) against a 60 kg reference adult. A 100 mg/kg daily mouse dose becomes a human equivalent dose near 8 mg/kg per day, about 490 mg daily — a safety-scaling number, not evidence that the resulting exposure is efficacious.

Can I just use the dose from the paper my protocol came from?

No — precedent alone is the one animal-dose origin the agent flags without exception, because a dose inherited through a citation chain carries no relationship to human exposure at any link in the chain. The defensible origins are exposure matching to a known human exposure, a body-surface-area or allometric conversion, the species’ maximum tolerated dose, or a stated pharmacodynamic target.

I dosed mice at 100 mg/kg a day — what does that mean in a human, and can I use it as evidence my drug will work?

A mouse dose of 100 mg/kg daily converts to a human equivalent dose near 8 mg/kg, about 490 mg daily for a 60 kg adult, using the FDA’s body-surface-area method. That number sets a safe starting dose for a first-in-human trial; it is not evidence the resulting exposure is efficacious, since the same guidance explicitly excludes exposure measurements from the starting-dose calculation.

Exposure has to be measured, not assumed

An administered dose is not an exposure, and the gap between them is where most in vivo translational claims fail quietly. Without measured plasma or tissue concentrations from the animals actually dosed, the achieved exposure is inferred from the dose, and that inference assumes bioavailability, clearance and distribution the study never measured. The agent reports the absence of any measured concentration as a finding in its own right, because it converts every downstream comparison from an observation into an assumption.

Three measurements settle it, and none requires a separate study. A sparse pharmacokinetic sample from satellite animals at two or three time points after the first dose gives a Cmax and a rough exposure. A terminal sample at the time the pharmacodynamic readout was taken links the effect to a concentration rather than to a dose. A tissue or tumour concentration at the same time point answers the question that plasma cannot: whether the compartment containing the target ever saw the concentration the in vitro experiment used. For central nervous system targets the relevant parameter is that same partition coefficient measured for brain, the unbound brain-to-plasma Kp,uu, because total brain concentration in a lipophilic compound routinely overstates free brain concentration by an order of magnitude. Here it usually runs the other way from metformin’s hepatic 15: brain Kp,uu can sit well below 1 even when the free drug principle holds cleanly for the rest of the body, because the blood-brain barrier’s own efflux transporters — P-glycoprotein and BCRP chief among them — actively pump many compounds back into blood; a manuscript claiming a CNS target needs the brain-specific coefficient and cannot borrow the liver or plasma free fraction as a substitute.

Choose the exposure metric the pharmacology requires. Cmax is the right comparator for a target that responds to peak concentration; area under the concentration-time curve is right for a cumulative effect; and the trough matters when the effect depends on continuous suppression. Reporting a single Cmax against an in vitro concentration that took 72 hours of continuous exposure to produce a phenotype is a mismatch of kind, not of magnitude — the cells sat at that concentration for three days, and the animal was above it for forty minutes. This is the same reasoning that makes a dosing schedule a substantive claim rather than a methods detail, and it is checked alongside in vivo experimental rigour and the ARRIVE reporting items for dose, route and timing.

Target engagement at the dose that produced the phenotype

Target engagement is the demonstration that the intended target was actually bound or inhibited in the animal, at the dose that produced the effect, and its absence leaves off-target activity as the standing alternative explanation. A phenotype at 100 mg/kg with no engagement measurement supports the sentence “this compound reduces tumour growth” and does not support the sentence “this compound reduces tumour growth by inhibiting kinase X.”

The measurement is usually cheap relative to the experiment that produced the phenotype. A pharmacodynamic marker downstream of the target — phosphorylation of a known substrate, an induced transcript, a metabolite ratio — sampled from the treated tissue at a defined time after dosing is the standard approach. Where no downstream marker exists, the cellular thermal shift assay described by Martinez Molina and colleagues in Science 2013;341:84–87 measures ligand-induced thermal stabilisation of the target protein directly in cells and tissue samples. For receptor-directed biologics, receptor occupancy by flow cytometry answers the question quantitatively.

The engagement question interacts with the exposure question in a way that is worth naming, because it is the one place where a high effective concentration can be defended. A compound that accumulates in the target tissue, or that binds its target irreversibly, can produce durable engagement at a plasma concentration far below its in vitro IC50 — and a manuscript that shows this has answered the objection rather than deflected it. The agent’s finding is not that a ratio is unacceptable; it is that a ratio has been left unstated where the reader will compute it. Where the mechanistic claim rests on two agents acting together, the same demand for a stated quantitative basis appears in drug synergy claims.

Frequently asked questions

What does target engagement mean, and do I actually need to prove it?

Target engagement is the demonstration that the intended target was bound or inhibited in the animal, at the dose that produced the effect — and yes, without it, off-target activity stands as the alternative explanation for the phenotype. A downstream pharmacodynamic marker, a cellular thermal shift assay, or receptor occupancy by flow cytometry are the usual ways to show it.

How do I answer a reviewer who says my concentration isn’t clinically relevant?

Give the ratio between the effective concentration and achievable exposure, name the free-versus-free comparison and its source, and quote the narrowed sentence you have already written into the abstract and discussion. A rewritten claim the exposure data support settles this comment far more often than an argument that the original concentration is defensible after all.

My in vitro results look great — will a reviewer still say my drug won’t work in real patients?

Often, yes, if the manuscript never compares the effective concentration to achievable human exposure — a clean dose-response curve and correct statistics do not answer that separate question. PerfectPaper’s translational dose agent runs that comparison itself, on a free-versus-free basis against reported human pharmacokinetic data, and reports the ratio next to the claim it qualifies before a reviewer has to.

Schedule: which pharmacodynamic index the compound obeys

The dosing schedule is a translational claim whenever the effect depends on time above a threshold rather than on peak concentration, and the manuscript has to say which index applies. Antimicrobial pharmacology is where this is most explicitly codified: beta-lactam efficacy tracks the fraction of the dosing interval during which free drug exceeds the minimum inhibitory concentration, aminoglycoside efficacy tracks the peak-to-MIC ratio, and glycopeptide efficacy tracks cumulative exposure. The 2020 revised consensus guideline of the American Society of Health-System Pharmacists, the Infectious Diseases Society of America, the Pediatric Infectious Diseases Society and the Society of Infectious Diseases Pharmacists recommends, for serious methicillin-resistant Staphylococcus aureus infections, “an individualized target of the AUC/MIC ratio of 400 to 600 (assuming a vancomycin MIC of 1 mg/L)”, the MIC determined by broth microdilution (Rybak et al., American Journal of Health-System Pharmacy 2020;77:835–864).

Oncology and immunology have the same structure with less standardised vocabulary. A once-daily mouse dose that produces a two-hour spike above the IC50 and a twenty-two hour trough below it is a different pharmacological experiment from a continuous infusion holding the same average concentration, and the two support different clinical schedules. Where a mouse study used a schedule with no clinical counterpart — five days on, two days off in a species with a fraction of the human half-life — the agent asks for the stated relationship rather than assuming one exists.

The agent brief

Paste this into a custom agent. Suggested settings: work type domain_review, skill level expert, tools scholar search.

Name: Translational dose relevance

You are reviewing whether the doses and concentrations used in a preclinical
study bear a defensible relationship to what is achievable in patients.

Determine and report:

1. Effective concentration. For every in vitro experiment supporting a
   mechanistic or therapeutic claim, identify the concentration at which the
   effect occurs. Report experiments where the concentration is not stated in
   the figure or legend.

2. Achievable exposure. Where the compound has been given to humans,
   determine whether the manuscript compares its effective concentration to
   reported human plasma exposure. Report the absence of any such comparison.
   Where the manuscript cites an exposure, check that the comparison is
   like for like: total against total, or free against free. Report
   comparisons of a total plasma concentration against a free in vitro
   concentration without an adjustment for protein binding, since this
   overstates achievable exposure, often substantially.

3. Magnitude. Where both numbers are available, state the ratio between the
   effective concentration and the reported achievable exposure. Where the
   effective concentration exceeds achievable exposure by more than roughly an
   order of magnitude, report this directly as the central limitation of the
   translational claim.

4. Animal dose basis. For in vivo experiments, determine how the dose was
   chosen: exposure matching to a human dose, allometric or body surface area
   conversion, maximum tolerated dose in that species, or precedent from prior
   literature. Report doses chosen by precedent alone. Report any conversion
   presented without naming its basis.

5. Exposure confirmation. Determine whether plasma or tumour concentrations
   were measured in the animal study. Report their absence, since without them
   the achieved exposure is assumed rather than known.

6. Target engagement. Determine whether engagement of the intended target was
   demonstrated at the dose producing the phenotype, by a pharmacodynamic
   marker or a direct measure. Report a phenotype attributed to a target where
   engagement was never shown, and note that off-target activity is the
   alternative explanation.

7. Schedule. Determine whether the dosing schedule bears a stated relationship
   to clinical use, particularly for agents whose effect depends on time above
   a threshold.

8. Language. Identify claims of clinical relevance, translational potential or
   therapeutic promise that are not supported by the exposure comparison, and
   propose specific rewording.

9. Route and formulation. Identify the route used in vivo and report any
   discussion of a clinical route that differs from it, particularly
   intraperitoneal dosing discussed as oral therapy. Report vehicles and
   formulations, including DMSO, cyclodextrin and nanoparticle delivery, where
   the discussion implies a conventional formulation.

10. Metabolites and compartment. Determine whether circulating metabolites
    were considered, and whether the concentration compared is from the
    compartment containing the target rather than plasma alone. Report a
    tissue-, tumour- or brain-resident target compared only against plasma.

Use scholarly search where enabled to locate reported human pharmacokinetic
parameters for named compounds. Cite what you rely on. Where no such data
exist, say so rather than estimating.

Editing the brief is expected, and adding a numbered item is the usual way to extend it. Antimicrobial reviewers add an item naming the pharmacodynamic index and the MIC distribution used; reviewers of biologics replace item 4 with a minimum anticipated biological effect level item; reviewers of natural-product work add an item on assay interference and compound stability in medium. How to write a custom reviewer agent covers the brief structure this one follows.

What this agent catches

How to find this in your own manuscript before a reviewer does

Run three passes over your own text and the objection surfaces in about thirty minutes. Translational dose defects are found by cross-referencing figures against the discussion, which is exactly the reading that stops happening after the fifth draft.

Pass one: build a two-column table of every concentration and dose in the paper. Left column, the number as it appears in each figure, legend and methods paragraph, with its unit and its exposure duration. Right column, the achievable human number with its source. Most manuscripts cannot fill the right column at all, and discovering that during drafting is the whole point of the exercise. Where a figure axis carries no unit, or a legend says “treated with compound X” and the concentration lives only in the methods, fix that first — the agent reports it, and so does a reviewer.

Pass two: search the abstract and discussion for the claim vocabulary. The retrieval handles are short and specific: clinically relevant, clinically achievable, translational, therapeutic potential, therapeutic window, promising, druggable, warrants clinical evaluation, patients. Every hit is a sentence that has to survive the ratio computed in pass one.

Pass three: check whether the number in the abstract is the number in the figure. Abstracts routinely quote the lowest concentration at which any effect was seen, while the mechanistic figures the claim rests on were run higher. A reviewer who finds a 1 micromolar abstract and a 20 micromolar western blot has found two problems, and the second one is harder to answer than the first.

What a reviewer writes when the dose is not achievable

Reviewers rarely write “translational dose relevance”. They write the specific version, and these are the sentences that appear in decision letters on preclinical therapeutics manuscripts.

“The concentrations used in Figures 2 and 3 are 10-50 micromolar; the reported Cmax in humans at the licensed dose is well below this range, and the authors should address whether the mechanism described is achievable in vivo.” “The comparison to human plasma levels does not account for plasma protein binding; the free concentration is the relevant comparator.” “The dose of 100 mg/kg in mice is not converted to a human equivalent dose, and the abstract describes the compound as clinically promising.” “No pharmacokinetic data are presented; the achieved exposure in the treated animals is unknown.” “Target engagement was not demonstrated at the effective dose, so the phenotype cannot be attributed to inhibition of the named target.” “The compound was administered intraperitoneally, and the discussion proposes oral administration in patients.” “The authors should state the ratio between the effective concentration and achievable exposure explicitly.”

The last one is worth anticipating, because it is the comment that arrives when the author has acknowledged the issue in general terms and given no number. A limitations paragraph saying that “translation to the clinic will require further study” reads as an admission without an analysis; a limitations paragraph saying that the effective concentration is approximately twenty-fold above the reported free Cmax, and that the work is therefore presented as mechanistic, reads as command of the pharmacology. Which of the two you wrote decides whether the comment comes back in the second round, and the limitations section is where the number belongs.

What to do when the effective concentration cannot be lowered

The narrower sentence is the remedy in most cases, and it is available immediately, without a new compound, a new experiment or new funding. A high effective concentration limits the translational claim; it does not invalidate the mechanistic finding, and the two are separable in a single afternoon of editing.

When the ratio is unfavourable and the biology is sound. Report the comparison in the results or discussion, state the ratio, and reframe the paper as mechanistic. “Compound X inhibits kinase Y with an IC50 of 8 micromol/L, approximately twentyfold above the free plasma concentration reported at the licensed human dose; the present work therefore characterises the mechanism rather than proposing the compound as a clinical candidate” is a sentence that survives review. The suppressed version of the same sentence does not.

When human pharmacokinetic data do not exist for the compound. Say so as a bounded fact rather than omitting the comparison. Name what is known — a structural class, a related compound with published exposure, an oral bioavailability measured in the same species — and state that human exposure has not been characterised. A stated absence is a specification; silence reads as an omission, and a reviewer will search for the data you did not cite.

When the target sits in a compartment you did not sample. Argue the compartment explicitly. Accumulation in tumour tissue, active transport into the target cell, irreversible binding and long target residence time are all legitimate reasons a plasma concentration understates target exposure, and each becomes an argument rather than a hope once a tissue measurement or a published partition coefficient supports it. Label the argument as correlative rather than causal where that is what the supporting data are.

When the reviewer asks for the pharmacokinetic study anyway. Answer in three parts: agree with the pharmacological point, quote the narrowed claim you have written in its place, and give a bounded reason the study is out of scope for this manuscript. A rewritten sentence quoted in the response letter settles this comment more often than an argument about the model does, and how to write a response to reviewers covers the structure.

How this differs from the built-in review

PerfectPaper’s translational dose agent differs from the standing review team in the kind of question it asks: not whether the evidence in the manuscript supports the claim as presented, but whether an exposure achievable in a human exists at all. The standing team’s conclusion-validity and methodology specialists ask the first question; this agent asks the second, and that comparison is what separates a mechanistic finding from a development argument — the reason many otherwise sound preclinical papers still draw a translational objection.

PerfectPaper reads every concentration and dose in the figures, legends and methods, sets them against reported human exposure on a free-versus-free basis, and reports the ratio next to the sentence in the abstract or discussion that depends on it. Pairs with the overclaim check, which catches the same sentences from the language side, and with model-to-human relevance, which asks whether the host and the site could support the claim at any dose.

Give this agent scholarly search. Unlike most agents in this collection, it needs external evidence to do its job. Full set: AI peer review for clinical trials, and the preclinical counterparts at AI peer review for preclinical cancer research.

Review my manuscript

Related agents

The agents below share a review with this one; you can run three at a time. See AI peer review for clinical trials for the full set of five therapeutics agents.

• Response criteria and endpoints — use it for confirmed responses, censoring rules and waterfall consistency.
• Early-phase safety and dose reporting — use it when the dose-limiting toxicity definition or its window is unstated.
• Registration and pre-specification — use it to catch a reported primary endpoint that differs from the registered one.
• Drug synergy claims — use it when synergy is asserted from a three-bar figure with no reference model.