Case Study: Factor B Inhibitor Profiling (Iptacopan)

Why this benchmark matters

Iptacopan is a high-bar retrospective test, on purpose. It combines an approved-drug label, a published human ADME record, a difficult Factor B / Factor D selectivity question that is driven by a specific structural motif, and enough public chemistry to test not just one molecule but rank-ordering across a real inhibitor panel — all without FluxMateria being trained on either target.

ADMET predictions versus the public record

Nine scored endpoints. Each FluxMateria output is compared directly to the published label or clinical ADME record. Pass means the prediction lands in the correct published class or envelope. Partial means directionally correct with a documented caveat, as in the plasma-protein-binding entry below.

Endpoint	FluxMateria prediction	Published reference	Status
Primary metabolic enzyme	CYP2C8 (top of panel)	CYP2C8, 98% (FDA label)	Pass
Metabolic soft spots	Benzylic oxidation sites flagged	Benzylic oxidation + UGT acyl glucuronidation	Pass
Reactive-metabolite liability	Acyl glucuronide alert fired automatically	M8 / M9 acyl glucuronide metabolites observed in the ADME study	Pass
Plasma protein binding	57% (single-point)	75–93% (concentration-dependent, target-mediated)	Partial
Brain barrier class	CNS− (peripheral)	Low, peripheral target	Pass
Oral absorption / permeability	Moderate Caco-2 class, oral bioavailable	Fa ≈ 71%, T_max 1.5 h (clinical ADME)	Pass
Aqueous solubility class	Low, formulation needed	XLogP 4.27; special formulation required for in-vivo work	Pass
Cardiac safety (hERG)	Low risk (IC₅₀ ≈ 16 µM)	IC₅₀ ≈ 415 µM, low risk (~230× above C_max)	Pass
Hepatic DILI	Moderate risk	Low to moderate; reactive metabolites flagged	Pass

What matters here: this is not a claim of decimal-place identity. It is a claim that nine scored endpoints landed in the correct published class or envelope, with zero hard failures, on a compound the platform was not tuned against. The one partial reflects a known structural limitation of single-point PPB prediction in the presence of target-mediated drug disposition.

Factor B versus Factor D selectivity

Selectivity is the hardest axis to call computationally, because the target and counter-screen are related serine proteases with overlapping structural features. Iptacopan is documented to be highly selective for Factor B over Factor D. Running the same molecule through FluxMateria against each target separately produced the following split:

Factor B (on-target)

Predicted pK_i 6.57

Directionally correct — the engine identifies iptacopan as a Factor B binder. Absolute affinity sits about 1.4 log off the published 10 nM IC₅₀, which remains inside the platform's stated triage-grade envelope for absolute single-compound potency.

Factor D (counter-screen)

Predicted pK_i 4.07

The output surfaces a specific mechanistic reason for the penalty: Factor D requires an amidine or guanidine warhead to engage its self-inhibited S1 pocket. Iptacopan carries neither, so the counter-screen is correctly pushed down.

Headline result

2.5 log-unit separation (~316×) between Factor B and Factor D, with an explicit structural rationale surfaced in the platform output. Recovering the selectivity class is the hardest single-line result on the page.

Rank-ordering on a 179-compound Factor B panel

A single approved-drug match is useful, but medicinal chemistry lives or dies on analog ranking. The full ChEMBL Factor B bioactivity panel (179 unique compounds; experimental IC₅₀ values spanning roughly 6.5 log units) was therefore used as a blind rank-ordering test. Zero of those 179 compounds were seen during development of the scorer, so the full panel is effectively one large hold-out.

Production scorer — full panel

Pearson r = 0.65, Spearman ρ = 0.67

Across all 179 compounds, the physics-derived scorer ranks with mean absolute log-error of 0.99 and median 0.83 log. Zero fitted parameters, zero training data for this target.

What earns the number

Binding-mode + geometry physics

The scorer is aware that Factor B's druggable pocket is allosteric rather than the catalytic S1, penalises orthosteric-warhead motifs on allosteric targets, and applies a gated carboxyl-to-aromatic geometry term on the iptacopan chemotype. No training, no fitting.

Per-scorer comparison on the 179-compound panel

0.65

Physics, binding-mode + geometry aware

Pearson r. Median |log-error| 0.83 log. Zero fitted parameters, zero training data. Default production path.

0.29

GPU 3D structural docking

Pearson r. Mean |log-error| 1.60. Kept for single-compound pocket analysis; mis-calibrates mid-potency.

0.14

Ligand-only physics (baseline)

Pearson r. Mean |log-error| 1.33. Reserved for per-compound energy decomposition; not a ranker.

Three scoring paths evaluated side by side. The binding-mode and geometry-aware physics engine is the production path for Factor B. The other two modes are kept as structural-analysis and interpretability tools on single compounds, and are reported here for transparency.

Discovery-series SAR: 16-compound iptacopan-lineage benchmark

The 179-compound panel above is breadth. This second benchmark is depth: a tighter, chemotype-homogeneous series that mirrors what a medicinal-chemistry team actually optimises against — peptide leads, the cyanobenzimidazole / indole scaffolds from the LNP023 discovery programme, iptacopan itself, and a related quinazoline-piperazine analog. Sixteen compounds spanning 4.3 log units of experimental IC₅₀.

Production scorer on the series

Pearson r = 0.70

Physics ranks the 16-compound series at Pearson 0.70 with mean absolute log-error 0.67 and median 0.63. Baseline ligand-only physics on the same series returns −0.38 Pearson (anti-correlated — it over-ranks peptides). The pocket-aware rule chain flips the sign.

Iptacopan itself

7.74 pred vs 8.00 exp

Within 0.26 log of the experimental pK_i on the flagship compound of the series. Peptide leads correctly penalised to ~5 pK_i; indole-cyanobenzimidazole analogs mostly land within 0.7 log of experiment.

What this second benchmark adds: on a chemotype-homogeneous SAR series — the kind of dataset a programme chemist would run at lead-optimisation stage — the same production physics chain holds up. No re-tuning between the broad 179-compound panel and the focused 16-compound series. Both are computed with the identical rule set.

The compound and the benchmark setup

Iptacopan (Fabhalta®, LNP023; Novartis) is the first-in-class oral small-molecule inhibitor of Complement Factor B, approved by the FDA in December 2023 for paroxysmal nocturnal hemoglobinuria. It is a particularly strong retrospective benchmark because the target is mechanistically clear, the ADME record is public, and the selectivity requirement against nearby protease space is real rather than cosmetic.

Iptacopan — input SMILES (public, PubChem CID 90467622) O=C(O)C1=CC=C([C@H]2N(CC3=C(OC)C=C(C)C4=C3C=CN4)CC[C@H](OCC)C2)C=C1

The benchmark question

Given only the SMILES and no label lookup, can FluxMateria recover the primary enzyme, the reactive-metabolite liability, the broad ADMET envelope, and the Factor B versus Factor D selectivity pattern that the public record reports for iptacopan?

Physicochemical baseline

All six basic physicochemical descriptors land inside the tolerance window of the published reference. This matters because a misleading physicochemical baseline can make every downstream ADMET success look accidental.

Property	FluxMateria	Published	Status
Molecular weight (g/mol)	422.5	422.5	Pass
TPSA (Å²)	74.7	74.8	Pass
logP	4.4	4.3	Pass
Hydrogen-bond donors	2	2	Pass
Hydrogen-bond acceptors	4	4	Pass
Rotatable bonds	7	7	Pass

How long it took

All timings below are wall-clock on the standard FluxMateria API rather than on a reserved GPU cluster. The point is not just speed for its own sake, but speed at the stage where medicinal chemistry still needs iterative feedback.

34 s

Full single-compound profile

All nine ADMET endpoints, physicochemical panel, pK_a, UV-visible features, Factor B and Factor D binding calls.

179 compounds

Full-panel rank-ordering

The full ChEMBL Factor B bioactivity panel through the physics binding path in a single batch, without per-compound tuning.

2 compounds

Additional paired profiling

Iptacopan and MGV354 profiled side by side under the same fixed settings in roughly a minute total.

What this page supports — and what it does not

Intended reading

This is a platform-transparency page, not a clinical claim about iptacopan. Iptacopan is already approved. The narrower claim is that FluxMateria can recover an approved drug's published profile from SMILES alone and rank analogs within the same target class at useful correlation strength.

What is production-grade

Single-compound ADMET profile

Nine scored endpoints, zero hard failures, wall-clock under a minute, and the reactive-metabolite flag fired automatically. This is the right mode for early candidate review.

What is production-grade

Target rank-ordering

Pearson 0.65 and Spearman 0.67 across 179 Factor B compounds spanning roughly six orders of magnitude of experimental IC₅₀, computed from physics alone with zero training data for this target and zero fitted parameters.

What is triage-grade

Absolute IC₅₀

Median absolute log-error is about 0.83 log across the full panel. Good for sub-micromolar vs micromolar calls and go / no-go triage. Not suitable for replacing a measured IC₅₀.

What is not claimed

Clinical outcome prediction

The page surfaces computational agreement and risk-profile structure. It does not claim that in-silico output alone can predict clinical success, trial outcome, or human translational biology.

Important limitation

All values reported here are computational predictions from FluxMateria. This page supports a platform-validation narrative. It does not make any new clinical claim about iptacopan and does not recommend any off-label use.

Research use notice: FluxMateria outputs shown here are not intended to diagnose, treat, cure, mitigate, or prevent disease, and must not be used for patient-specific clinical decision-making. Any therapeutic claim would require experimental validation, additional preclinical work, regulatory review, and where applicable controlled clinical trials.

Additional retrospective comparison: iptacopan versus MGV354

This paired comparison is included as a secondary context block rather than as the main proof. Its purpose is to test whether the broader risk profile converges in the right direction on a later clinical failure, while being explicit that the actual human failure mechanism was not a pure ADMET issue.

Metric	Iptacopan (approved)	MGV354 (Phase 2 failure)
Plasma protein binding	57%	99.4% (f_u ≈ 0.6%)
DILI risk class	moderate	high
hERG IC₅₀	16.4 µM (Low)	6.0 µM (Moderate)
Intrinsic clearance	99.5 µL/min/mg	347 µL/min/mg (~3.5× faster)
Brain barrier class	CNS−	CNS−

Important caveat: FluxMateria would not have called the actual clinical failure mechanism for MGV354. The value of the comparison is narrower: even when the human failure was driven by species translation rather than metabolism alone, the in-silico risk profile still tilts in the expected direction on several independent axes.

Selected references

Public sources behind every number on this page

Every reference value in the comparison tables is sourced from public clinical, regulatory, or chemistry databases. No internal data, no private assay results, no proprietary chemistry set.

FDA label, Fabhalta® (iptacopan) 218276s000lbl, 2023.
PubChem CID 90467622; IUPHAR/BPS ligand 10710.
Clinical ADME study (iptacopan human mass-balance), 2024.
Discovery medicinal-chemistry paper for LNP023 (J. Med. Chem., 2020).
ChEMBL bioactivity target CHEMBL5731 (Complement Factor B), full IC₅₀ panel used for the blind rank-ordering test.
MGV354 discovery (J. Med. Chem., 2018) and Phase 1/2 trial (Ophthalmology, 2018).

Talk to us about how this applies to your chemistry

The most useful next step is a 30-minute technical walkthrough with your chemistry team on a target of your choice. We can also share the public JSON packet — headline metrics, predicted-vs-published ADMET, physicochemical comparison, Factor B / Factor D selectivity annotations, SAR correlation results, and the paired-compound comparison.

Request a 30-minute walkthrough Download public JSON

FluxMateria recovered iptacopan's 2.5-log Factor B / Factor D selectivity from SMILES alone — with an explicit structural reason, and no training data for either target.