DDIA HEALTH ONE
/ RESEARCH WORKSPACE
Company
DIA RESEARCH / REPRODUCIBILITY SERIES 01

Make the evidence
reproducible.

A computational research workspace examining glucose-responsive insulin and paracetamol glucuronide. We reanalyse published measurements, check the small-molecule setup, and show what is still needed before target binding simulations.

Explore the evidence ↓Download source values ↗
01 EVIDENCE FIRST COMPUTATIONAL NON-CLINICAL
01 / PUBLISHED BENCHMARK

Glucose response, reanalysed.

NNC2215 · hIR-A receptor affinity
relative to human insulin
Measured response and exploratory Hill fit
Values from the Nature 2024 source workbook · Figure 3d
3.16×mean paired change
3 → 20 mmol/L
NNC2215 hIR-A affinity source data at five glucose concentrations, showing three independent experiments, mean and standard deviation, an exploratory shared-slope Hill curve, and a wide experiment-bootstrap range.
Open points show the three source experiments; bars show mean ± sample SD. The fitted curve is descriptive, and the green band resamples the three experiments. Its upper EC50 tail reaches the fit ceiling.
Source data reanalysis · relative affinity (% of human insulin)
Glucose (mmol/L)NNC2215 mean ± SDInsulin degludec control mean
Loading source-data summary…
DESCRIPTIVE MODEL / IDENTIFIABILITY CHECK

The fitted midpoint is weakly identified.

EXPLORATORY
APPARENT HILL SLOPE1.12Bootstrap range 0.67–1.49
EC50 · mmol/L13.497.5th percentile reaches the 1,000 mmol/L fit ceiling

The model shares EC50 and slope across experiments, fixes each measured zero-glucose baseline, and fits a separate upper response per experiment. Only three experiments are available; the upper bootstrap tail is censored, so no reliable midpoint or 6 mmol/L threshold is established.

Fit and provenance JSON ↓Reproduce fit ↗Source-only chart ↗
ASSAY hIR-A · 1.5% HSA · REPLICATES 3 independent experiments
MOLECULAR PREMISE / PUBLISHED MODEL

Two linked elements. A glucose-sensitive equilibrium.

Conceptual schematic
not to scale
Conceptual NNC2215 glucose-sensitive switch The 2024 paper proposes that, at low glucose, a tethered glucoside can occupy the B29 glucose-binding macrocycle, favoring a closed-like state. Free glucose can bind the macrocycle and shift the equilibrium toward an open-like state. This is a conceptual diagram, not an atomic structure. LOW FREE GLUCOSE GLUCOSE PRESENT INSULIN CORE Phe B1Lys B29 B29 MACROCYCLE G Tethered O1-glucoside occupies macrocycle Closed-like state favoured Open-like state favoured INSULIN CORE Phe B1Lys B29 B29 MACROCYCLE G G Free glucose competes for macrocycle GLUCOSECOMPETITION
LOW FREE GLUCOSE
INSULIN COREPhe B1Lys B29GB29 MACROCYCLE

Tethered O1-glucoside occupies the macrocycle; closed-like state favoured.

GLUCOSE PRESENT
INSULIN COREPhe B1Lys B29GGB29 MACROCYCLE

Free glucose competes for the macrocycle; open-like state favoured.

ARCHITECTURE B29 Lys Nε: glucose-binding macrocycle attached through a linker and triazole. B-chain PheB1: short-linker O1-glucoside. MODEL STATUS Publisher-supplied open and closed NNC2215 PDB models are now catalogued below. They are model-built coordinates, not experimentally resolved NNC2215 structures; this diagram remains conceptual.
REFERENCE STRUCTURESPDB 6PXV ↗PDB 2VJZ ↗Used as modelling references in the paper; neither is an experimentally solved NNC2215 switch-state structure.
PUBLISHER-SUPPLIED COORDINATES · INTERACTIVE 3D

Inspect the open and closed model-built snapshots.

Drag to rotate and scroll to zoom. Each view reads the original PDB coordinates; the color palette separates insulin, the B29 macrocycle, and the B1 glucoside.

Loading structures
Insulin chains A / B B29 macrocycle B1 glucoside Free glucose in open model
OPEN STATEPublisher model
990 atoms
Free glucose is recorded as chain D. No insulin receptor is included.
CLOSED STATEPublisher model
968 atoms
No free glucose is recorded in this model. No insulin receptor is included.
MODEL STATUS

These are author-supplied, model-built coordinates, not experimentally solved NNC2215 switch states or simulation outputs. The display does not validate atom mapping, bond orders, stereochemistry, protonation, charges, or force-field parameters. Use the download links to inspect the source files.

Open model PDB ↓Closed model PDB ↓Source checksums ↓Viewer version & license ↓Publisher source ↗
02 / METABOLITE INTERACTION

Paracetamol glucuronide is a distinct molecule.

The 2026 study reports stronger NNC2215 binding to PaG than to glucose, with increased receptor binding at low glucose. PaG is a paracetamol metabolite; the parent drug is not the tested species.

View the 2026 study ↗
REPORTED DISSOCIATION CONSTANT · Kd
D-GLUCOSE1.91 mM± 0.10 mM
VS
PaG0.17 mM± 0.01 mM
11.2×lower reported Kd for PaG, calculated from the reported means
ZERO-GLUCOSE hIR-A AFFINITY · REPORTED
NO PaG0.9%
→
70 μM PaG5.4%
6.0× from rounded source means1.5% HSA assay
FIGURE 3 / DATA AVAILABILITY

The published effect is clear; the point-level dataset is not.

The paper reports the zero-glucose means in text and shows the broader glucose response as a figure.

Open source audit ↓
01 / ARTICLE TEXT 0.9% → 5.4% 0 mmol/L glucose · no PaG vs 70 μmol/L PaG EXACT TEXT VALUES
02 / FIGURE 3 Mean ± SD hIR-A assay · 1.5% HSA · three independent experiments VISUAL FIGURE
03 / SOURCE TABLE Not supplied No numeric curve table, exact SD values, or replicate data in the reviewed article and SI. NO CURVE REFIT
SOURCE BOUNDARY The paper reports p < 0.0001 from a two-way ANOVA; this project did not recalculate it. The public demo links the CC BY-NC-ND 4.0 article instead of copying or digitizing Figure 3. Open the paper ↗SI inventory ↓
IN VIVO / PIG GLUCOSE CLAMP

More glucose was needed to hold the clamp at 3 mmol/L.

GIR AUC · 300–420 min
reported mean ± SEM · mg/kg
TreatmentPublished AUC summarynP vs salineRatio to saline*
Low-dose PaG 0.5 g
684.3 ± 107.5
80.161.15×
High-dose PaG 2.0 g
849.0 ± 178.5
70.041.42×
L-glucose 0.3 g/kg
790.1 ± 120.2
80.011.32×
Saline control
596.8 ± 85.9
8—1.00×
* DERIVED Ratios divide published group means by the saline mean; they are descriptive, not paired animal-level effects. P-values and SEM are reported by the paper. Low-dose PaG was not statistically significant versus saline. Published Table 1 ↗Download clamp values ↗Model parameters ↗Derived analysis JSON ↗Calculation script ↗
MODEL REPRODUCTION / PUBLISHED EQUATION 4B

Reapply the reported PaG sensitivity coefficient.

SI = SIbase × (1 + PaG × CoeffPaG / 100)

0.7784%SI increase per μmol/L PaGfitted pig PK/PD parameter · CV 5.5%
Human · 1 g paracetamolestimated mean maximum PaG · 70 μmol/L
+54.5%1.54× baseline SI
Pig · 0.5 g IV PaGreported mean maximum PaG · 102 μmol/L
+79.4%1.79× baseline SI
Pig · 2.0 g IV PaGreported mean maximum PaG · 426 μmol/L
+331.6%4.32× baseline SI
INTERPRETATION Equation-derived change in the fitted insulin-sensitivity parameter. This is not receptor affinity, a glucose-infusion prediction, or an individual-patient outcome. Article & model source ↗Concentration inputs ↗Result JSON ↗Calculation script ↗
ASSAY FINDING In vitro receptor-binding data; the reported 70 μM condition is not a patient-specific exposure estimate. DATA LIMIT PaG figure-level raw values were not available for independent recalculation in this demo.
ORIGINAL SIMULATION / REPORTED IN SUPPORTING INFORMATION

The published MD recipe is recovered; its seed coordinates are not.

METHOD TRANSCRIBED
500 nsproduction phase
Desmond 6.6Schrödinger 2021-2
OPLS4force field
SPCexplicit water
0.15 MNaCl
STARTING INPUT An in-house crystal structure of a related glucose-bound macrocycle; the article says its coordinates are not shown. The public NNC2215 state models below are a different structural input. Open PaG Supporting Information ↗Download protocol record ↓
WAITING ON STUDY TEAM · EXACT-RERUN INPUTS

The handoff is ready; the missing pieces are specific.

The published run used a glucose-bound macrocycle analogue and a trajectory-derived PaG start. We have prepared a compact request list for the files that would make the original setup reproducible.

AWAITING SOURCE FILES
01 / SYSTEMCrystal coordinates + topologyOriginal glucose-bound analogue, component definitions, atom mapping, and Desmond system files.
02 / PaG STARTExact frame + build projectTrajectory frame/time used for the PaG pose and the Maestro project or script that built the starting system.
03 / RE-RUNParameters + settings + enginePaG protonation/charges, full run configuration, and access to the reported Desmond 6.6 / OPLS4 stack.
REPRODUCTION GATE Validate received chemistry and inputs before launching MD. A simulation from the public 2024 NNC2215 models would be a separate proxy study. No target binding MD has started. Open full scientist handoff ↗
03 / REPRODUCIBILITY RUN

Re-run the source analysis.

4 published-data reanalyses · 1 analogue PMF reanalysis
source values checked against original records
ANALYSIS READYINPUT · FIGURE 3D CSV

Replicate-level fold change

Recalculate group means, sample standard deviations, and paired fold changes from the deposited CSV. The browser run hashes its input and creates a downloadable JSON manifest.

Download latest result ↓
COMPUTE: CPUINPUT CHECKSUM: Available on runTARGET BINDING MD: NOT RUNCLINICAL MODEL: NONE
View Python analysis ↗Read source & methods ↗
CURRENT RUN REGISTER

Four published-data reanalyses, one Hill-fit check, a PaG ligand-file audit, and a public analogue PMF reanalysis complete. Target binding MD has not started.

RUN 02 · ACS 2026REPRODUCED

PaG pig clamp · Table 1

1.147×Low dose vs saline
1.423×High dose vs saline

Derived ratios of published group means. Individual animal data and statistical tests were not recomputed.

Run JSON ↓Python script ↗
RUN 05 · NATURE 2024WEAKLY IDENTIFIED

Hill response fit · Figure 3d

n = 1.12apparent slope
13.4 mMbest-fit EC50

Bootstrap over three experiments reaches the 1,000 mM fit ceiling for EC50. This is an exploratory fit, not a reliable midpoint or activation threshold.

Fit JSON ↓Python script ↗
RUN 06 · PAG LIGAND FILE AUDITFILE AUDIT COMPLETE

GAFF2 connectivity + stereochemistry

39 / 39atoms in source order
5 / 5CIP labels preserved

All 40 bonds map to the source graph. OpenMM loads the Amber files with net charge −1.000001 e and finite initial energy. This checks PaG alone, not conjugate parameters or binding.

Audit JSON ↓Audit script ↗GAFF2 Mol2 ↓
RUN 07 · TARGET MOLECULAR DYNAMICSNOT STARTED

Macrocycle analogue · PaG

500 nspublished production method

Published method: Desmond 6.6, OPLS4, SPC water and 0.15 M NaCl. A 500 ns glucose-bound macrocycle-analogue trajectory supplied the snapshot used to build the PaG start; the same protocol was then applied. The in-house crystal coordinates and trajectory are not public, so an exact rerun is blocked. Our NVIDIA runner is a separate PaG-in-water portability check, not binding MD, and remains unrun.

Published method ↓Proposed pilot ↓Latest solvent pilot ↓NVIDIA runner source ↗Run instructions ↗Readiness gates ↘
PUBLIC MODEL AUDIT · RUN MANIFEST AVAILABLE

113 heavy atoms match across states; source formulas localize the count gap to the B29-linked pool.

Both publisher models match their recorded hashes, desB30 sequences, backbone coverage, three native disulfides, and documented modification links. The selected PHX/LYX/MC heavy-atom graph has 120 matching CONECT/LINK edges after two explicit LYX name reconciliations. Source-reported B1/B29 building blocks reproduce the published formula difference; the model inventory still has two extra carbon records, with atom mapping and chemistry unresolved.

BOND ORDER / CHARGE OPEN
2 / 2source hashes match
113heavy atoms · both states
120 / 120adjacency edges match
0conjugate force fields validated
HYDROGENS & PARAMETERS OPEN

The graph comparison normalizes LYX CD6 ↔ C (the C31-linked roof atom) and OD ↔ O (the O atom with CD1 as its only heavy-atom neighbor). Hydrogen records still differ: the closed model assigns the nearby PheB1 N-terminal H to PHX without a CONECT edge and has one fewer recorded LYX CD2 hydrogen neighbor. Adjacency does not establish bond order, protonation, formal charge, stereochemistry, or force-field parameters. These are model-built coordinates and include no insulin receptor.

PUBLISHED FORMULA · SOURCE COMPONENT COUNTS

Source chemistry narrows the two-carbon record gap.

FORMULA ARITHMETIC · NOT AN ATOM MAP
PUBLISHED NNC2215C₃₂₅H₄₆₁N₇₉O₉₇S₆full formula · Nature SI
DESB30 INSULIN COREC₂₅₃H₃₇₆N₆₄O₇₅S₆sequence + 3 disulfides
EXPECTED TOTAL ATTACHMENTC₇₂H₈₅N₁₅O₂₂source formula difference
PDB PHX / LYX / MCC₇₄N₁₅O₂₂heavy-atom records
B1 · PHEB1 GLUCOSIDE / PHXCOUNT MATCH
Short-linker glucoside
Source-derived attachmentC₁₀H₁₆O₈
PDB heavy atomsC₁₀O₈

Element counts agree after excluding hydrogen records.

B29 · LYS Nε MACROCYCLE / LYX + MC+2 C RECORDS
Macrocycle, linker and triazole
Source-derived attachmentC₆₂H₆₉N₁₅O₁₄
PDB heavy atomsC₆₄N₁₅O₁₄

Carbon records differ by two; nitrogen and oxygen counts agree.

B1 + B29 source-derived attachmentsC₇₂H₈₅N₁₅O₂₂=Published NNC2215 − desB30 coreC₇₂H₈₅N₁₅O₂₂

The PHX count matches the B1 component; the combined LYX/MC inventory is two carbons above the source-derived B29 component. Counts narrow the discrepancy to this residue pool but do not identify an atom or validate a chemical graph. The open model’s missing oxygen record is ASN A21 OXT; the closed model includes it. Bond orders, stereochemistry, protonation, charges and force-field parameters remain unvalidated.

Nature synthesis and formula source ↗Patent Example 2 cross-check ↗
Structure audit ↓Validation script ↗Formula decomposition ↓Formula audit script ↗Milestone report ↗Proposed pilot plan ↓
LOCAL COMPUTE / ENGINE PREFLIGHT

OpenMM OpenCL is running on the Apple M4.

A generic, solvated ubiquitin benchmark checks that the local simulation stack can build and run a protein system.

LOCAL ACCELERATOR · NO NVIDIA
Single-run throughput for GROMACS CPU, OpenMM CPU and OpenMM OpenCL on a 26,821-atom ubiquitin system. OpenMM OpenCL on Apple M4 measured 126.274 nanoseconds per day; this is not an NNC2215 or NVIDIA run.
One 5,000-step run per engine on the same 1UBQ system. The chart is an installation and throughput preflight, not a controlled head-to-head benchmark.
26,821solvated atoms · 1UBQ
5,000timed steps · 2 fs
4CPU threads
0NVIDIA GPUs used
How to read this

Each bar is one short throughput run. The engines use different thermostat implementations and timing boundaries, so the chart is an environment preflight, not a controlled engine ranking. It does not validate the PaG force field, reproduce the published 500 ns method, or establish binding energetics.

Run manifest ↓Re-run script ↗Pinned environment ↗Benchmark input PDB ↓RCSB source ↗
NVIDIA / OPENMM GPU PREFLIGHT

OpenMM OpenCL ran on a Tesla T4.

Three independent 100 ps timing windows on a generic, solvated 1UBQ system measured a median throughput of 270.2 ns/day.

FREE COLAB T4 · RUN COMPLETE
Three OpenMM 8.6.1 OpenCL 100-picosecond throughput measurements on a Tesla T4 using a generic 17,764-atom ubiquitin system. Median 270.2 nanoseconds per day. No PaG or NNC2215 is included.
Three independent velocity seeds; OpenMM 8.6.1 used the OpenCL platform on the Tesla T4. This generic benchmark is not a target-system throughput estimate or an NNC2215–PaG simulation.
270.2median ns/day · 3 runs
257.9–282.7replicate range · ns/day
17,764atoms · 1UBQ
15 GBTesla T4 GPU memory
Interpretation

This is a real NVIDIA GPU engine preflight through a temporary free Colab runtime. Its force field and system differ from the Apple M4 run; do not treat the two rates as a controlled comparison. It does not validate NNC2215 conjugate parameters or predict PaG binding.

GPU run manifest ↓Re-run on NVIDIA OpenCL ↗Rebuild chart ↗Pinned OpenMM version ↗Benchmark input PDB ↓
LOCAL RESEARCH / PAG-ONLY SOLVATION PILOT

A small-molecule setup completed two short explicit-water runs.

Two independent 100 ps OpenMM runs used the modeled PaG glucuronate in water and salt on Apple M4 OpenCL.

SOLVENT ONLY · NO BINDING
Temperature and box-volume time series from two 100-picosecond PaG-only explicit-water OpenMM runs. Samples fluctuate around 298 kelvin; no NNC2215 or binding partner is present.
The shaded first 10 ps are equilibration; the remaining 100 ps are the short production interval. The traces check setup behavior, not conformational equilibrium or binding.
2 × 100 psindependent velocity seeds
1,272atoms · PaG + water + ions
298.9 Kreplicate 1 production mean
297.5 Kreplicate 2 production mean
Interpretation

Both production traces remained finite and had mean temperatures near the 298.15 K target. The periodic box contains one PaG at an implied concentration of about 0.118 M; this is a finite-box consequence, not the 70 μM assay condition. No NNC2215, macrocycle, receptor, binding energy, or NVIDIA GPU is included.

Latest run pointer ↓Run manifest ↓Production summary ↓Replicate 1 trace ↓Replicate 2 trace ↓Re-run pilot ↗Rebuild figure ↗PaG identity & checksum ↓Ligand parameter record ↓Locked environment ↗
PUBLIC METHOD ANALOGUE / SOURCE-TRAJECTORY REANALYSIS

WHAM reanalysis lands within the published GluHUT uncertainty.

We reprocessed the authors’ public umbrella-sampling trajectories with Grossfield WHAM 2.0.11. This is a methods check on a biomimetic receptor analogue.

ANALYSIS COMPLETE · CPU WHAM
Three independent GluHUT and glucose unbinding free-energy profiles, with a mean and between-run standard deviation. The reproduced mean binding free energy is minus 12.78 plus or minus 1.02 kilocalories per mole, compared with the published minus 12.5 plus or minus 0.9.
PMFs are reanalysed from deposited trajectories. The binding free energy is the global profile minimum relative to the 13–15 Å bulk region; shaded band shows between-run ±1 sample SD.
−12.78 ± 1.02reanalysed ΔG · kcal/mol
−12.5 ± 0.9published ΔG · kcal/mol
3 × 51PMFs × windows per PMF
0new MD runs · CPU post-processing
SCOPE LIMIT

GluHUT is a biomimetic glucose receptor; these files contain D-glucose, not PaG or NNC2215. This reanalysis checks a public WHAM workflow and does not establish NNC2215–PaG binding, receptor activity, or clinical effects.

Reanalysis JSON ↓Re-run WHAM analysis ↗Source article ↗Source trajectories ↗Pinned WHAM source ↗
04 / SIMULATION READINESS

What has to be true before binding MD?

No guessed chemistry.
No unvalidated simulation.
01
COMPLETE

Source data

Nature Figure 3d values are transcribed with source sheet and cell references. Paired experiment ratios reproduce the published rounded glucose-response benchmarks. PaG binding ratios are recalculated from the article's rounded means.

30 CELLS · SHA-256 TRACKED
02
CANDIDATE ATOM MAP · CHEMISTRY STILL OPEN

Structure & parameters

The Nature supplement’s B1 and B29 building-block formulas imply C72H85N15O22 for the attachments. Public model records contain C74N15O22. A candidate mapping from the numbered synthesis schemes and PDB connectivity localizes the two-carbon residual to the MC residue; it does not identify the individual surplus atoms or establish a validated chemical topology. The open-state oxygen difference is the missing ASN A21 OXT record. Bond orders, stereochemistry, protonation, charges and conjugate parameters remain unresolved.

PUBLISHER SESSION / FIGSHARE · CC BY 4.0

Open and closed snapshots match the published PDB records.

CHECKSUM VERIFIED
ACTIVE / OPENMODEL-BUILT
990 atoms
54 residues 1,010 bond edges
INACTIVE / CLOSEDMODEL-BUILT
968 atoms
53 residues 986 bond edges

Atom identity sequence matches the publisher PDBs.
The open snapshot matches after the documented BGLC → BGL residue-name alias.

Bond ordersNOT STORED
Explicit atom typesNOT PRESENT
Formula residual+2 C IN MC*

*Under the candidate map: source C3–C8 follow the six-carbon B29 linker; C21–C23 follow the three-carbon macrocycle tail. The individual surplus MC atoms remain unidentified. The session also lacks authoritative bond orders, explicit target-residue atom types, and validated force-field parameters.

Candidate map audit ↓Re-run mapping ↗Session audit JSON ↓Audit script ↗Source linker scheme ↗Source macrocycle scheme ↗Original 79 MB CXS ↗
Open-state PDB ↓Closed-state PDB ↓Checksums & provenance ↓Structure audit ↓Serial connectivity audit ↓Formula decomposition ↓Synthesis & component audit ↓Macrocycle reference audit ↓Candidate linker map ↓Re-run map audit ↗2026 SI file audit ↓Proposed pilot ↓
MODEL-BUILT · CANDIDATE MAP ONLY · PARAMETERS UNVALIDATED
03
6 NVIDIA CATALOG REQUESTS PENDING · T4 PREFLIGHT COMPLETE

NVIDIA compute access

Inception portal rechecked 18 Oct 2026 at approximately 10:00 UTC: all six partner-benefit requests remain pending; no GPU allocation is shown. The Products page lists two Dia Health records (both Developing / Not Accelerated), satisfying NVIDIA’s published one-product prerequisite. Explore Benefits opened the public catalog; a completed “Innovation Lab” catalog search returned 0 results, so the documented Request Benefit step is unavailable and no application was submitted. NVIDIA describes this as a selective 60-day program and lists molecular dynamics, protein folding and drug-discovery pipelines among supported simulation use cases. The current T4 evidence remains a generic 1UBQ preflight; no PaG portability or target-binding run has started. A HunAIFA support intake was submitted for Hungarian/European AI Factory guidance; no reply or compute allocation is confirmed. Partner-program receipts are not awards or compute allocations.

Lambda · AWS · Microsoft Azure$1,000 · $10,000 · $5,000 · partner applications incomplete
PENDING
AI Factory FranceUp to 50K GPU hours · partner application not submitted
PENDING
Scaleway€40,000 · Inception request pending; partner application incomplete
PENDING

Eligibility gates: AI Factory France requires France-based R&D and open research with publication; a cross-border referral has been requested through HunAIFA but remains unconfirmed. The two visible product records do not establish an active NVIDIA GPU workload. No award or GPU allocation is confirmed. AI Factory France requirements ↗

Other GPU offers screened: Nebius lists up to $5,000 in credits plus future GPU savings, but requires VC/accelerator backing and at least $5 million raised; the current profile does not meet that threshold. NVIDIA says Scaleway reviews its offer and a separate partner application may be required. Nebius criteria ↗ · Scaleway partner application ↗

European benchmark route: EuroHPC Development Access may support OpenMM/GROMACS code testing and optimization. EuroHPC Fast Lane is intended for AI activities; the current workload is classical MD. The free T4 was a generic 1UBQ preflight only; no NNC2215–PaG target simulation has run.

Open Inception benefits ↗Innovation Lab requirements ↗Inception support ↗HunAIFA support ↗EuroHPC Development Access ↗Application brief ↗
6 CATALOG REQUESTS PENDING · HUNAIFA INQUIRY SUBMITTED · NO GPU AWARD · TARGET MD NOT STARTED
i
Why the Hill midpoint remains unresolved

The descriptive fit uses five glucose concentrations from three independent experiments. Its EC50 bootstrap upper tail reaches the 1,000 mmol/L numerical fit ceiling, showing that these data do not identify a reliable midpoint. The apparent Hill slope is not a binding-site count, and this analysis does not establish a 6 mmol/L switch threshold.

View fit record ↗
SOURCE REGISTER

Every number has a source.

Published evidence, extracted values, derived calculations, and missing inputs are kept distinct.

01Høeg-Jensen et al.Nature · 2024 · NNC2215 architecture and glucose response↗ 02Nature Source Data · Figure 3dOriginal workbook · NNC2215 and insulin degludec source values↗ 03Sturis et al.ACS Pharmacol. Transl. Sci. · 2026 · PaG interaction↗ 04PaG articlePublished receptor-binding and pig-clamp findings↗ 05PaG Supporting InformationMD recipe, model parameters, and supplementary figures · SI DOI 10.1021/acsptsci.6c00069.s001↗ 06Published NNC2215 open/closed model PDBsJohansson & Granata · Springer Nature Figshare · CC BY 4.0 · checksums in manifest↗ 07NVIDIA Innovation Lab60-day GPU access for selected Inception members · portal request required · rolling review↗ 08Google Cloud Free Trial restrictionsGPU attachment and quota requests require a paid billing account↗ 09Nature Supplementary InformationNNC2215 synthesis route, B1/B29 modifications, and reported molecular formula↗ 10Tromans et al.Nature Chemistry · 2019 · related glucose-macrocycle model; SI reports methods and figures, no coordinate file in the reviewed PDF↗ 11PaG SI file auditACS Figshare API record · single PDF and checksum verified · no coordinate or trajectory files listed↓ 12NNC2215 serial-connectivity auditPublisher PDB CONECT/LINK records · duplicate-name mapping, resolved links, and remaining bond-order gaps↓ 13wwPDB PDB connectivity specificationCONECT records identify atom serials; charge distinction and bond orders require separate chemical evidence↗ 14PaG receptor-assay source auditExact text values, CC BY-NC-ND 4.0 license, and missing machine-readable curve data↓ 15Eades et al. · GluHUT glucose receptorChemical Science · 2026 · public host–guest PMF analogue and umbrella-sampling method↗ 16GluHUT umbrella-sampling datasetZenodo · deposited windows, trajectories, and WHAM inputs used in this reanalysis↗ 17PaG GAFF2 ligand-only auditSource graph, CIP stereochemistry, charge, and OpenMM Amber-file load check↓ 18AI Factory France eligibilityR&D location, open-research and publication requirements↗ 19EuroHPC Development AccessHPC application testing, porting, and optimization access call↗ 20Google for Startups Cloud ProgramPublished eligibility tiers; Google makes final award decision↗ 21Macrocycle lineage audit2024 roof-azide scheme and cited 2019 receptor SI; provenance improves, NNC2215 topology remains unvalidated↓ 22NNC2215 formula decompositionSource-reported B1/B29 components; two extra carbon records localized to the B29-linked PDB pool↓ 23Novo Nordisk patent WO2023144240A1Example 2 names the B1/B29 conjugate and reports calculated ions matching the 2024 supplement; no atom-mapped PDB topology↗ 24Independent browser reproduction manifestLive Figure 3d run, 18 Oct 2026 09:34 UTC · input checksum matches Python · no GPU simulation↓ 25ChimeraX session structure auditPublisher session checksum, model inventory and atom-identity comparison · chemistry gate remains open↓ 26ChimeraX session serialization sourcePinned source used to interpret which atom types and bond fields the CXS session stores↗ 27Candidate NNC2215 linker and tail mapNumbered Nature schemes matched against PDB CONECT paths; +2 C localized to MC under this inferred partition↓ 28Archived Python source-data rerun18 Oct 2026 09:58 UTC · 30 Nature workbook cells independently verified · source and script checksums recorded↓ 29Latest Python source-data rerun18 Oct 2026 11:15 UTC · same source and script checksums · paired fold changes unchanged↓

RESEARCH DEMONSTRATION · NOT A CLINICAL MODEL · NOT FOR TREATMENT OR INSULIN-DOSING DECISIONS. This page reanalyses selected published in-vitro source data and reports selected pig-clamp summary values. Its short PaG-only solvent simulations do not reproduce a binding study, establish safety, or predict an individual’s response.