RN Collins (Series 3) No.6:Tiers, Targets, and the Guidance Gap: A Proposed Evidentiary Framework for Computational Pharmacology in FDA Botanical Drug Proceedings

RN Collins (Series 3) No.6

RN Collins has written a new series (3) of 18 articles for cannabis law report on 2026 Psychedelics & Legal Issues.

This is the 5th in the series

Contact RN Collins: https://www.linkedin.com/in/rn-collins/

Tiers, Targets, and the Guidance Gap: A Proposed Evidentiary Framework for Computational Pharmacology in FDA Botanical Drug Proceedings

RN Collins, M.S., JD Candidate

https://www.linkedin.com/in/rn-collins/

ABSTRACT

Computational pharmacology has produced a pipeline for identifying protein targets of bioactive compounds at a scale and speed that would have been difficult to imagine when FDA’s 2016 Botanical Drug Development Guidance was written. Reverse docking methodologies, corroborated in some cases by molecular dynamics simulation and macromolecular crystallography, now generate multi-target profiles for cannabinoids that appear to have no established evidentiary status in FDA proceedings. This may reflect a regulatory gap with three dimensions, though each warrants some caution before being treated as settled. First, FDA’s substantial evidence standard, calibrated to a single-target model of pharmacology, does not appear to provide a formal framework for evaluating computational target data in botanical drug submissions. Second, citizen petition responses that dismiss computational pharmacology evidence without meaningful explanation could raise arbitrary-and-capricious concerns under APA § 706(2)(A), a question that has taken on added weight after Loper Bright, though how a reviewing court would actually resolve such a challenge is far from certain. Third, structure/function claims in cannabis labeling that are built on computational data without adequate qualification may create exposure under FTC substantiation standards. This Article proposes a four-tier evidentiary framework, distinguishing computational prediction (Tier 1), molecular dynamics validation (Tier 2), in vitro structural confirmation (Tier 3), and in vivo confirmation (Tier 4), and applies it, tentatively, to FDA botanical drug submissions, FRE 702 expert qualification, and APA arbitrary-and-capricious review. It closes by sketching what a computational pharmacology annex to the 2016 Guidance might look like, and suggests that such an annex could help close the regulatory gap and reduce an apparent arbitrage in which unsupported claims proliferate while companies that invest in rigorous computational science receive no clear evidentiary credit for doing so. These are proposals, not predictions, and the framework offered here is meant as a starting point for discussion rather than a finished doctrinal solution.

I. Introduction

Multi-target pharmacology appears to be something close to the default operating mode of plant-derived bioactive compounds, though “default” may overstate how uniform the underlying biology actually is. The canonical single-target drug model, one compound, one receptor, one measurable clinical outcome, describes a synthetic pharmaceutical design ideal that most natural products do not seem to approximate closely. Cannabis sativa produces over one hundred distinct cannabinoids, each structurally complex enough to plausibly interact with multiple protein binding sites, though the extent of that interaction in any given cannabinoid is not always well characterized. The single-target framing of cannabinoid pharmacology, THC as a CB1/CB2 agonist, CBD as a compound with poorly characterized mechanisms, has long functioned as a simplification. What may have changed is that validated computational and structural methods now make that simplification harder to sustain as a matter of regulatory or legal argument, though “harder to sustain” is not the same as “untenable,” and reasonable regulators could disagree about how much weight the newer data should carry.

The difficulty is, at bottom, evidentiary. FDA’s drug approval framework appears to have been designed around the single-target model: a compound with a defined mechanism acts on a defined receptor to produce a defined clinical outcome, measured in adequate and well-controlled investigations satisfying the substantial evidence standard.¹ Multi-target compounds may resist this framework in practice: they have many potential mechanisms, potentially acting at different targets under different dosing conditions and in different tissues. Clinical trials that measure a clinical outcome cannot always identify which of a compound’s many mechanisms produced it, and it is not clear that existing trial designs are well suited to disentangling that question even where the underlying pharmacology is multi-target.

The problem has become more acute in the cannabinoid context for a specific reason that deserves some hedging of its own: the emergence of computational pharmacology methodologies that can generate multi-target profiles for cannabinoids at a scale that was not previously practical, and that appear, on the available evidence, to have some degree of validation. A 2019 study published in ACS Omega described a reverse docking pipeline that identified 151 protein targets for acteoside, 143 for quercetin, and 128 for EGCG, while recovering a majority of experimentally confirmed interactions for each compound.² The 2023 EMBL crystallography study reported what appears to be the first crystal structure of THC bound to autotaxin, a secreted enzyme implicated in cancer, fibrosis, and inflammation.³ Taken together, these studies may represent an evidentiary category that FDA has not yet formally engaged, though it is possible the agency has considered and simply not published its views on the subject, and this Article does not claim to have full visibility into internal agency deliberations.

This Article considers the possible legal implications of this scientific development across three bodies of law that intersect in FDA proceedings: (1) FDA’s evidentiary standards for botanical drug submissions; (2) the Federal Rules of Evidence standards that might govern expert testimony on computational pharmacology data in federal litigation; and (3) the administrative law framework for APA review of FDA decisions touching computational pharmacology evidence after Loper Bright. The through-line of the analysis is a four-tier evidentiary framework that attempts to sketch the legal weight that each type of computational and experimental evidence might reasonably carry in each of these contexts. The framework is presented in Part III and applied, with appropriate qualification, throughout.

Part II describes the computational pharmacology science at a level of detail intended to make the tier distinctions intelligible. Part III introduces the four-tier framework in tabular form. Part IV applies the framework, tentatively, to FDA botanical drug submissions. Part V addresses FRE 702 and Daubert analysis for computational pharmacology expert testimony. Part VI addresses APA review after Loper Bright. Part VII sketches what a formal FDA computational pharmacology annex might contain. Part VIII concludes, with appropriate caveats about the speculative nature of some of the proposals.

II. The Science: Reverse Docking, Molecular Dynamics, and Structural Confirmation

A. The Computational Pipeline

Reverse docking, also called inverse docking or target fishing, inverts the standard structure-based virtual screening paradigm. Where standard docking screens a library of compounds against a defined protein cavity, reverse docking screens a library of protein structures for cavities that appear to accommodate a defined ligand. The ACS Omega 2019 pipeline illustrates the methodology: Zhang et al. used pharmacophore comparison through the PharmMapper server to filter a protein structure database to a more manageable candidate pool, then applied AutoDock Vina for binding energy calculation at a threshold of −8.5 kcal/mol.⁴

The pipeline’s validation is probably the evidentiary bedrock of any Daubert analysis, though it is only a partial one. Of 24 predicted ingredient-target complexes selected for comparison against known PDB structures, 16 (67%) reportedly showed highly similar binding modes, 9/9 for acteoside, 6/7 for quercetin, 5/8 for EGCG. On its face, the pipeline appears to recover a majority of experimentally confirmed interactions, at least within this limited validation set. Its false positive rate for predicted targets that are not subsequently confirmed remains unknown at scale, and this uncertainty is probably the principal Daubert vulnerability, addressed further in Part V. Nothing about this pipeline should be read as establishing that any particular predicted target is real; it may only suggest which targets are worth further investigation.

B. Molecular Dynamics Simulation as Refinement

The ACS Omega study used 100-nanosecond molecular dynamics simulation to refine docking predictions for three selected complexes: NOS2 (iNOS) with acteoside, PDE5A with quercetin, and WARS with EGCG. Simulation was implemented in GROMACS with the AMBER-14SB force field, in systems of roughly 60,000 to 140,000 total atoms.⁵ Binding free energy was calculated using the g_mmpbsa tool: acteoside-NOS2 yielded −264.145 kcal/mol; quercetin-PDE5A, −185.516 kcal/mol; EGCG-WARS, −55.807 kcal/mol.

Molecular dynamics simulation appears to serve a distinct and possibly legally significant function: it converts a docking score, a single number representing a predicted binding affinity under idealized, static conditions, into something closer to a dynamic picture of the ligand-protein complex under simulated physiological temperature and pressure. A predicted complex that proves unstable under MD simulation would likely provide weaker evidence than one that remains stable across replicated runs, though how much weaker is a matter of scientific judgment rather than fixed legal rule. This distinction may justify treating MD-validated evidence as a separate tier, though reasonable scientists could draw the line differently.

C. Structural Confirmation: The EMBL Autotaxin Case

Macromolecular crystallography can provide structural confirmation at a resolution that computational methods alone cannot achieve: something approaching atomic-level visualization of the ligand-protein complex in the crystal state. The EMBL Grenoble study by Eymery, McCarthy & Hausmann (2023) reported three-dimensional crystal structures of THC bound to autotaxin (PDB: 7P4J and 7P4O), along with what the authors described as confirmed inhibition in vitro.⁶ The researchers also reported statistically significant inhibition for 6a10aTHC, THCA, CBD, and S-DMH-CBD at P < 0.005 for ATX-β and ATX-γ isoforms, which may extend the finding across the broader cannabinoid class, though extrapolation beyond the specific compounds tested should probably be treated with some caution.

Notably, the researchers characterized the evidentiary status of their own finding with considerable care: in vitro structural confirmation, with in vivo significance still pending. Their observation that “in vivo studies are necessary to confirm that the binding between autotaxin and THC is linked to the therapeutic effects of THC administration,” because “the main known targets of THC in the human body are the CB1 and CB2 cannabinoid receptors,” seems to place the current state of the evidence at roughly Tier 3. This caveat, offered by the researchers themselves, may be legally significant, since it appears to mark a rough boundary between claims that are more readily defensible and claims that are not.

III. The Four-Tier Evidentiary Framework

The following framework attempts to distinguish four levels of computational and experimental evidence, each associated with a tentative evidentiary weight in FDA proceedings, FRE 702 analysis, and APA judicial review. The framework is offered as the central contribution of this Article; the remaining analysis applies it, with the understanding that any such taxonomy involves some degree of simplification and could reasonably be drawn differently by others.

Tier

Evidence Type

May Establish

Arguably Sufficient For

Likely Not Sufficient For

Cannabis Example

FDA Submission Stage

1

Computational prediction (reverse docking score)

A hypothesis-generating target candidate

IND Phase I background; mechanistic rationale

Structure/function claims; mechanism representations; efficacy claims

ACS Omega pipeline: 151 targets identified for acteoside

IND Stage 1 background only

2

Molecular dynamics validation (stable simulation, binding free energy)

A predicted interaction that appears stable under simulated physiological conditions

IND Phase I mechanism section (qualified); GRAS notice pharmacology section

Efficacy claims; unqualified mechanism representations in labeling

ACS Omega MD validation: acteoside-NOS2, quercetin-PDE5A, EGCG-WARS

IND Stage 1-2; qualified claims only

3

In vitro structural confirmation (crystallography, cryo-EM, SPR, ITC)

Binding that appears confirmed at atomic resolution in vitro

A qualified mechanism claim (“research suggests … in laboratory studies”); IND Phase I-II mechanism section; structure/function claim with in vitro qualification; citizen petition factual submission

Efficacy claims; clinical-level mechanism representations; unqualified “works through” language

EMBL Autotaxin: PDB 7P4J/7P4O (THC-autotaxin crystal structure, 2023)

IND Stage 1-2; NDA background; qualified claims arguably supportable

4

In vivo confirmation (PK/PD studies, target engagement, ex vivo)

A pharmacological effect that appears confirmed in living organisms at therapeutically relevant concentrations

Efficacy claims; NDA mechanism-of-action labeling; unqualified mechanism representations

—

Not yet reported for the autotaxin-THC interaction; in vivo studies apparently pending

NDA/BLA required for approval-stage mechanism claims

The framework is not intended as a novelty so much as an attempt to articulate distinctions that experienced pharmacologists may already draw informally, and that FDA reviewers may apply somewhat inconsistently, in part because no formal framework currently requires consistency. Making the distinctions explicit, and tentatively connecting them to legal consequences that might flow from each tier, is the modest contribution this Article aims to offer.

IV. The FDA Framework: Botanical Drug Guidance and Its Computational Pharmacology Gap

A. The Substantial Evidence Standard

The FDCA requires “substantial evidence” of effectiveness for drug approval, defined as “evidence consisting of adequate and well-controlled investigations, including clinical investigations, by experts qualified by scientific training and experience.”⁷ This standard, developed in what might be called the single-target era, may create an evidentiary gap for multi-target compounds: on one reading, it requires that a compound’s clinical effect be established before it becomes clear which of its many targets produced that effect. A cannabinoid that inhibits autotaxin, modulates CB1/CB2, and interacts with computational targets identified by reverse docking presents something close to this problem, though the statute itself does not speak directly to multi-target compounds, and this reading is an inference rather than a certainty.

B. The 2016 Botanical Drug Guidance

FDA’s 2016 Botanical Drug Development Guidance appears to acknowledge some limitations of the single-target model for plant-derived drugs.⁸ The guidance permits botanical drug sponsors to characterize active constituents without complete mechanistic understanding, to rely on historical use data, and to use standardized extract characterization rather than single-compound mechanism data. What the guidance does not appear to address is computational pharmacology as a distinct evidentiary category, likely because the guidance was written at a time when reverse docking pipelines of the ACS Omega variety existed mainly as academic tools and had not yet achieved the validation status, throughput, or regulatory visibility that might have required FDA to engage with them directly.

The result may be a regulatory gap, though its practical significance is difficult to quantify from the outside. A sponsor who submits computational pharmacology data in a botanical drug IND appears to have no guidance on how FDA will weigh it. A sponsor who omits such data, while making structure/function claims that are in tension with computational data already in its own files, could face FTC scrutiny over the adequacy of substantiation. Neither outcome seems obviously well-reasoned as a matter of policy. A formal computational pharmacology annex to the guidance might help address both concerns, though it would not necessarily resolve every difficulty.

C. The Citizen Petition Context

The computational pharmacology gap may be particularly acute in the citizen petition context.⁹ Cannabis industry participants have used citizen petitions fairly extensively to request FDA action on cannabinoid regulatory classification. The apparent legal problem is that FDA does not appear to have an established standard for weighing computational pharmacology data in citizen petition responses. An agency response that dismisses computational pharmacology evidence without meaningful explanation could be vulnerable to challenge under APA § 706(2)(A) as arbitrary and capricious, though whether any particular petitioner could actually mount a successful challenge would depend heavily on the specific facts and the agency’s stated reasoning.¹⁰ An agency response that relies on computational data without assessing its tier-level quality might be similarly vulnerable, for related reasons. The four-tier framework proposed in this Article is offered as one possible analytical structure that could help FDA make more consistent and judicially defensible decisions on computational pharmacology evidence in citizen petition proceedings, though adoption of any such framework would, of course, be the agency’s choice.

D. Applying the Tiers to Botanical Drug Submissions

The four-tier framework may map, at least loosely, onto the botanical drug IND/NDA pathway, though the mapping below is offered as a proposal rather than a description of settled practice.

At IND Stage 1 (Phase I safety), Tier 1 and Tier 2 evidence would appear sufficient to populate a mechanism-of-action section. FDA does not generally require mechanistic certainty for a Phase I IND; what is typically required is enough mechanistic understanding to design a reasonably safe first-in-human protocol. Computational target profiles and MD-validated interactions may help identify potential pharmacological interactions relevant to dose selection and safety monitoring, though they would not establish that any such interaction is clinically significant.

At IND Stage 2 (Phase II efficacy), it may be reasonable to expect Tier 3 evidence to accompany a target-specific efficacy claim. Where a sponsor represents that a cannabinoid product works through autotaxin inhibition, in vitro structural confirmation of that interaction would seem to be a sensible evidentiary predicate before a Phase II trial is designed around it, though FDA has not adopted this as a formal requirement and sponsors may reasonably take different views on how much confirmation is enough. Absent Tier 3 confirmation, the Phase II hypothesis may rest on a thinner evidentiary foundation than the guidance seems to contemplate.

At NDA/BLA (approval), something closer to Tier 4 evidence would likely be expected for any target-specific mechanism claim in approved labeling. The substantial evidence standard for efficacy generally requires clinical investigation, which in turn typically involves some form of in vivo target engagement. On the evidence reviewed for this Article, the autotaxin pathway for cannabinoids does not appear to have reached Tier 4 as of this writing.

V. Federal Rules of Evidence 702 and Computational Pharmacology Expert Testimony

A. The Daubert Framework Applied

Federal Rule of Evidence 702 requires that expert testimony be based on sufficient facts or data, be the product of reliable principles and methods, and reflect reliable application of those methods to the facts of the case.¹¹ The Daubert inquiry directs courts to consider whether the methodology has been tested, peer-reviewed, has a known error rate, and is generally accepted in the relevant scientific community.¹²

The reverse docking methodology used in the ACS Omega study appears to satisfy several of these criteria, at least on a preliminary reading. The pipeline is built on AutoDock Vina, a fairly widely validated and published molecular docking software package. The pipeline was validated against known ligand-target interactions with a reported 67% recovery rate across 24 test complexes, a modest sample that may limit how far the validation can be generalized. The study is peer-reviewed. The principal Daubert vulnerability is probably the unknown at-scale false positive rate, the rate at which predicted targets turn out not to be confirmed in vitro or in vivo. A court that conflates Tier 1 computational prediction with Tier 3 structural confirmation risks either over- or under-weighting the evidence, and the four-tier framework is offered as a possible structure that could help a court draw these distinctions with somewhat more precision, though it is ultimately for courts, not commentators, to decide how much weight any given category of evidence deserves.

B. The Expert Qualification Question

Courts applying FRE 702 to computational pharmacology expert testimony may face a qualification question that does not typically arise with more traditional experimental pharmacology: the methodological steps in a reverse docking pipeline appear to call on different expertise at each stage. Pharmacophore comparison seems to call for computational chemistry expertise; molecular dynamics simulation for structural biology expertise; crystallographic structure determination for macromolecular crystallography expertise; and in vivo pharmacokinetics for pharmacology expertise more generally. It seems unlikely that a single expert would have deep expertise across all four areas, though this is not a certainty and some experts may have unusually broad backgrounds. This may argue for either a multi-expert approach or a generalist expert able to provide an integrated overview of the methodology while deferring to the specialized literature on individual steps. Litigants navigating this area may find it useful to consider the tier level of the evidence at issue before expert designations are finalized, since the tier level may bear on what kind of expertise is actually needed.

C. The Hearsay Problem and Published Pipelines

Computational pharmacology pipelines rely on public databases, the Protein Data Bank for protein structures, ChEMBL and PubChem for ligand-target interaction data, KEGG for pathway analysis, that were not created for purposes of litigation. Expert reliance on these databases may implicate the hearsay rules in ways that traditional experimental data typically does not.¹³ The ACS Omega study’s reliance on the Protein Data Bank to obtain reference protein-ligand complexes for validation, and on the PharmMapper server for pharmacophore comparison, illustrates the issue reasonably well. Each database could be viewed as a form of hearsay if the underlying data is offered to prove the truth of the matter asserted, though this characterization is not free from doubt. FRE 703 permits experts to rely on otherwise inadmissible facts or data where experts in the field reasonably rely on that type of data in forming their opinions, which would seem to support reliance on well-established databases like the PDB, though it may not extend as readily to reliance on less established computational databases absent additional foundation.

VI. Administrative Law: Post-Loper Bright Review of FDA Computational Pharmacology Decisions

A. The Loper Bright Shift

The Supreme Court’s decision in Loper Bright Enterprises v. Raimondo, 144 S. Ct. 2244 (2024), overruled Chevron U.S.A. Inc. v. Natural Resources Defense Council, 467 U.S. 837 (1984), eliminating the presumption of deference to agency interpretations of ambiguous statutes.¹⁴ Courts must now exercise their own judgment in determining the meaning of statutory text, though how much practical difference this makes in any particular case may vary considerably. The potential impact on computational pharmacology evidence in FDA proceedings can be considered at two levels, each somewhat speculative.

At the statutory interpretation level, if FDA were to interpret the substantial evidence standard to exclude computational pharmacology evidence on the ground that the statute requires “adequate and well-controlled investigations,” a court reviewing that interpretation would now need to determine independently whether the statute actually mandates this exclusion, and it is possible, though far from certain, that a court could conclude it does not, on the theory that computational investigation of a compound’s target profile is itself a form of investigation even if it is not a clinical investigation in the traditional sense.

At the evidentiary standard level, if FDA’s guidance documents were to articulate a standard for evaluating computational pharmacology evidence, that standard would now appear to be subject to independent judicial review as a question of whether it faithfully implements the statutory mandate, rather than being treated as a policy choice automatically entitled to deference. Whether courts would in fact scrutinize such guidance this closely remains to be seen.

B. The Arbitrary and Capricious Standard

Even before Loper Bright, FDA decisions touching computational pharmacology evidence were arguably reviewable under APA § 706(2)(A) as arbitrary and capricious if they failed to consider relevant factors, relied on factors Congress did not intend the agency to consider, or failed to provide a reasoned explanation.¹⁵ Motor Vehicle Manufacturers Association v. State Farm Mutual Automobile Insurance Co., 463 U.S. 29, 43 (1983), held that an agency “must examine the relevant data and articulate a satisfactory explanation for its action including a ‘rational connection between the facts found and the choice made.'”¹⁶

FDA’s apparent failure to develop a formal standard for evaluating computational pharmacology evidence in botanical drug submissions may mean that its current citizen petition responses and IND-related guidance lack a clearly demonstrated “rational connection” between the computational pharmacology data submitted by petitioners and the agency’s evaluation of that data. An FDA response that dismisses Tier 3 evidence (confirmed in vitro structural interaction) with the same explanation applied to Tier 1 evidence (computational prediction alone) could be vulnerable to an arbitrary-and-capricious challenge under both the State Farm standard and the somewhat heightened scrutiny that Loper Bright may now invite, though, again, whether any particular challenge would succeed cannot be predicted with confidence.

C. The Major Questions Doctrine

The major questions doctrine, articulated in West Virginia v. EPA, 597 U.S. 697 (2022), requires clear congressional authorization for agency actions of “vast economic and political significance.”¹⁷ The doctrine could plausibly apply where FDA sought to use computational pharmacology evidence to make a classification decision, for instance, determining that a cannabinoid’s multi-target profile brings it within the drug definition, that affects an industry valued in the tens of billions of dollars. An agency asserting authority to make such a classification on the basis of computational pharmacology evidence, without a formal evidentiary framework and without clear congressional authorization, could conceivably face a major questions challenge, though this remains a somewhat speculative application of the doctrine to a fact pattern that has not yet been tested in court. The four-tier framework proposed in this Article is offered, in part, as an attempt to reduce this risk: it tries to specify the evidentiary showing that might reasonably be expected before FDA treats computational pharmacology data as dispositive in a classification decision.

VII. Toward a Formal FDA Computational Pharmacology Annex

A. What the Annex Might Contain

FDA and other relevant regulators may wish to consider supplementing the 2016 Botanical Drug Guidance with a computational pharmacology annex addressing the following elements. These are offered as recommendations for regulatory consideration, not as claims about what the law currently requires.

1. Methodology requirements. FDA could specify minimum standards for reverse docking pipelines used in botanical drug submissions, including the protein structure database scope, the scoring function and threshold, the MD simulation parameters that might be expected for Tier 2 validation (for example, a minimum of 100 ns, a documented force field, and binding free energy calculation), and documentation of false positive rates where available. The ACS Omega study may offer a useful methodological benchmark, though FDA would presumably want to consult more broadly before adopting specific parameters.
2. Tiered submission requirements. FDA could specify which tier of evidence might reasonably be expected at each stage of the IND/NDA pathway, roughly consistent with the framework proposed in Part III: Tier 1 evidence for background sections; Tier 3 evidence before Phase II efficacy trials target a specific computationally identified interaction; and Tier 4 evidence for NDA mechanism-of-action labeling.
3. Disclosure obligations. FDA could clarify when computational pharmacology findings might warrant disclosure outside the formal IND/NDA submission context, for example, whether a structurally confirmed novel target interaction (Tier 3) should be reported under the IND safety reporting obligations of 21 C.F.R. § 312.64, a question the current regulations do not appear to answer directly.
4. Citizen petition standards. FDA could establish standards for weighing computational pharmacology evidence in citizen petition proceedings, including what tier of evidence might be needed to compel a substantive agency response, and the grounds on which FDA could appropriately decline to engage with computational pharmacology evidence without addressing its specific technical claims.

B. The Industry Incentive Structure

The apparent absence of a formal FDA computational pharmacology standard may create a kind of regulatory arbitrage, though the scale of this effect is difficult to measure. A cannabis company that uses computational pharmacology to identify a novel cannabinoid target interaction, and then makes structure/function claims based on that interaction without formally submitting the computational data to FDA, may receive little regulatory credit for the underlying research while still assuming whatever liability risk attaches to the claim. A company that does formally submit the data may, at present, receive no clear guidance on how FDA will weigh it. A computational pharmacology annex could help address this asymmetry by creating a clearer submission pathway in which disclosure is incentivized and evidentiary credit is at least roughly commensurate with evidence quality, though designing such an annex well would require considerably more input from FDA scientists and industry stakeholders than this Article can offer.

VIII. Conclusion

The computational pharmacology methods described in this Article appear to have developed somewhat faster than the regulatory framework surrounding them, though it is possible FDA is further along internally than public materials suggest. The ACS Omega reverse docking pipeline, the EMBL autotaxin crystallography, and a number of similar studies in the cannabinoid literature may together represent an evidentiary category that FDA has not yet formally engaged in public guidance. If so, the result may be a regulatory gap that tends to benefit companies making less well-supported claims while offering little formal recognition to companies that invest more heavily in the underlying science, though this Article does not claim to have quantified that effect, and the picture could look different with more data.

The four-tier framework proposed in this Article, distinguishing computational prediction (Tier 1), MD validation (Tier 2), in vitro structural confirmation (Tier 3), and in vivo confirmation (Tier 4), is offered as one possible piece of doctrinal infrastructure that FDA might draw on in considering a computational pharmacology annex to the 2016 Botanical Drug Guidance. The framework may also be useful, in a more modest way, to litigants and courts working through FRE 702 challenges to computational pharmacology expert testimony, and to those assessing APA arbitrary-and-capricious claims against FDA decisions that arguably ignore or misweigh this category of evidence.

The underlying trend seems likely to continue. Cannabis rescheduling and increasing investment in cannabinoid drug development may well produce a growing number of IND submissions, citizen petitions, and litigation matters in which computational pharmacology evidence plays some role. Whether the framework proposed here proves useful in that context remains to be seen, but it seems reasonable for the legal and regulatory community to begin thinking through these questions before, rather than only after, the relevant cases and submissions accumulate.

ENDNOTES

1. Federal Food, Drug, and Cosmetic Act § 505(d), 21 U.S.C. § 355(d) (defining “substantial evidence” as “adequate and well-controlled investigations”), https://www.law.cornell.edu/uscode/text/21/355.
2. Haiping Zhang et al., Large-Scale Target Identification of Herbal Medicine Using a Reverse Docking Approach, 4 ACS Omega 9710 (2019), https://doi.org/10.1021/acsomega.9b00020.
3. Mathias C. Eymery, Andrew A. McCarthy & Jens Hausmann, Linking Medicinal Cannabis to Autotaxin–Lysophosphatidic Acid Signaling, 6 Life Sci. Alliance e202201595 (2023), https://doi.org/10.26508/lsa.202201595.
4. Zhang et al., supra note 2, at 9710–11 (describing PharmMapper server pharmacophore comparison and AutoDock Vina binding energy calculation at −8.5 kcal/mol threshold).
5. Id. at 9712–13 (describing 100-nanosecond GROMACS simulations with AMBER-14SB force field and g_mmpbsa binding free energy calculation).
6. Eymery et al., supra note 3; Mylène André, New Molecular Insights into Medical Cannabis, PHYS.ORG (Jan. 9, 2023), https://phys.org/news/2023-01-molecular-insights-medical-cannabis.html.
7. 21 U.S.C. § 355(d), supra note 1.
8. FDA, Botanical Drug Development: Guidance for Industry (Dec. 2016), https://www.fda.gov/media/93113/download.
9. 21 C.F.R. § 10.30 (governing citizen petition procedures), https://www.ecfr.gov/current/title-21/chapter-I/subchapter-A/part-10/subpart-C/section-10.30.
10. Administrative Procedure Act § 706(2)(A), 5 U.S.C. § 706 (authorizing courts to set aside agency action that is “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law”), https://www.law.cornell.edu/uscode/text/5/706.
11. Fed. R. Evid. 702, https://www.law.cornell.edu/rules/fre/rule_702.
12. Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579, 592–95 (1993) (setting out the factors for assessing reliability of expert methodology), https://supreme.justia.com/cases/federal/us/509/579/.
13. Fed. R. Evid. 703 (permitting experts to rely on otherwise inadmissible data if experts in the field reasonably rely on that type of data).
14. Loper Bright Enters. v. Raimondo, 144 S. Ct. 2244, 2261 (2024), https://www.supremecourt.gov/opinions/23pdf/22-451_7m58.pdf.
15. 5 U.S.C. § 706(2)(A), supra note 10.
16. Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43 (1983), https://supreme.justia.com/cases/federal/us/463/29/.
17. West Virginia v. EPA, 597 U.S. 697, 723 (2022) (requiring “clear congressional authorization” for agency actions of “vast economic and political significance”), https://supreme.justia.com/cases/federal/us/597/20-1530/.