Case Study: Alzheimer's Amyloid-Pathway Candidate Discovery

Why This Is Novel

It did not win by rediscovering symptomatic chemistry.

The study was explicitly structured to reject cholinergic fallback through an AChE anti-target. The final leaders stayed inside a root-cause BACE1 lane and still produced two independent chemotypes, not a single fragile scaffold story.

Why This Is Valuable

FluxMateria already did the early discovery compression.

The platform did not stop at finding possible binders. It discovered, ranked, counter-screened, and prioritized the field down to candidates and near-misses that are already medicinal-chemistry usable. The strongest failures are narrow permeability or hERG misses, which means the next move is targeted optimization and lab validation, not blind search.

Why This Is Faster

Potency, anti-target control, and CNS triage happened in one workflow.

Instead of splitting discovery across separate potency, safety, and counter-screen tools, FluxMateria returned a ranked and interpretable shortlist inside a single campaign, with two full confirmation reruns completed in under twenty-four minutes.

The challenge

Alzheimer's discovery programs often blur together two very different goals: symptomatic control and disease-modifying intervention. FluxMateria was not asked to rediscover symptomatic cholinergic molecules. It was asked to move upstream into amyloid-pathway biology and still keep the chemistry inside a realistic CNS safety envelope.

For this case study, BACE1 was used as the root-cause discovery axis because it sits upstream of amyloid production. To prevent the workflow from claiming success by drifting back toward symptomatic chemistry, AChE was encoded as an explicit anti-target. KCNH2 and hard CNS ADMET filters were used to remove candidates that looked potent on paper but would not survive early safety triage.

The question

Can FluxMateria generate novel BACE1-directed Alzheimer's candidates that preserve CNS feasibility, avoid explicit cholinergic rediscovery, and survive strict hERG and permeability gates without relying on a GPU refinement pass?

Proposed mechanism of action

This study is not just saying these molecules bind a target. It is making a specific mechanistic hypothesis about how they could help. The intended path is upstream amyloid-pathway suppression through BACE1 inhibition, not symptomatic cholinergic boosting.

APP Amyloid precursor protein is available for processing in neuronal biology.

BACE1 Cut Beta-secretase cleavage opens the amyloidogenic route this study is trying to suppress.

Amyloid-beta Output Downstream processing can generate amyloid-beta peptides once the BACE1 step occurs.

Therapeutic Hypothesis Inhibit BACE1 upstream, reduce amyloid-beta production pressure, then validate experimentally.

01

BACE1 sits early in amyloid-beta generation.

APP can be processed down an amyloidogenic route in which BACE1 performs the beta-secretase cut. That cleavage is the upstream entry point for amyloid-beta production.

02

After BACE1 cleavage, downstream processing can yield amyloid-beta.

Once APP is cut by BACE1, the resulting fragment becomes the substrate for subsequent processing that generates amyloid-beta peptides. Those peptides are central to the amyloid-pathway hypothesis in Alzheimer's disease.

03

These shortlisted molecules are intended to reduce that upstream flux.

The candidates in this page were selected because FluxMateria predicts strong BACE1 inhibition while keeping them inside a CNS-feasible and early-safety-filtered envelope. The intended effect is less amyloid-beta production pressure, not direct symptomatic neurotransmitter enhancement.

04

This is different from symptomatic AChE drugs.

AChE inhibitors can improve cholinergic signaling and temporarily help memory and cognition symptoms, but they do not directly target the same upstream amyloid-production step. That is why AChE was treated as an anti-target in this case study.

What is still unproven: this case study does not show plaque clearance, clinical benefit, or patient response. The mechanistic hypothesis still needs biochemical confirmation of BACE1 inhibition and cell-based confirmation that the shortlisted molecules actually reduce amyloid-beta output.

Literature context

This result is notable because FluxMateria did not converge on arbitrary Alzheimer's chemistry. In 23m 28s of confirmed local reruns on a home PC, it landed in the same two scaffold families that anchored major BACE1 programs: lanabecestat-like and verubecestat-like chemistry.

The family match is meaningful

The top shortlist is not merely amyloid-directional in a generic sense. It clusters around two medicinal-chemistry families that large pharmaceutical programs already treated as serious BACE1 starting points.

There is real preclinical precedent

Published work on lanabecestat showed that this chemical neighborhood can produce potent, orally active, brain-penetrant BACE1 inhibitors with strong amyloid-beta biomarker effects. That makes the surviving FluxMateria motifs biologically credible, not speculative.

Clinical caution still applies

Verubecestat and lanabecestat programs showed that the class can engage the target and move amyloid biomarkers, but they did not deliver clear clinical success in Alzheimer's trials. So this page supports a strong preclinical discovery claim, not a solved therapeutic claim.

Why FluxMateria still matters

The value here is speed and discrimination. FluxMateria reached this historically credible scaffold neighborhood quickly on local hardware, then filtered within it under explicit AChE, KCNH2, and CNS ADMET pressure to produce a smaller, more defensible assay package.

What this means: FluxMateria did in minutes what normally takes much longer to assemble by hand: it navigated into a medicinal-chemistry region that prior BACE1 programs had already shown to be serious, then separated the cleaner candidates from the historically familiar liabilities. That is exactly the kind of in-silico leverage that can make the next experimental cycle sharper and cheaper.

What was and was not pre-fed: FluxMateria was explicitly given a BACE1-centered amyloid discovery brief, anti-target constraints, CNS ADMET gates, and a target-appropriate search space. It was not given lanabecestat, verubecestat, or an instruction to reproduce known pharma series. Those families emerged from the broad search. After that point, the workflow deliberately ran chemotype-specific refinement sweeps and fed the validated motifs back into the production reruns, which is exactly how a real medicinal-chemistry campaign should behave.

Selected sources: Lanabecestat preclinical medicinal chemistry; Verubecestat trial in mild-to-moderate Alzheimer's disease; Lanabecestat neuroimaging follow-up; 2024 review of BACE1 inhibitors.

Exact FluxMateria brief

The final evidence set was generated from a strict discovery specification. This is the exact logic that produced the published shortlist.

Desired target

BACE1
Mechanism: inhibitor
Hard gate: pKi >= 6.8

Anti-targets

AChE pKi <= 7.0
KCNH2 pKi <= 5.0
Both enforced as hard constraints.

CNS ADMET

BBB penetrant
PPB <= 0.95
Permeability >= -6.0
hERG pIC50 <= 5.0
Hepatotox risk <= medium

Discovery scope

Broad: amyloid-oriented BACE1 discovery space
Native-only: BACE1-native scaffold confirmation space
Max enumeration: 1500
Top-ranked results retained: 25

Why include AChE as an anti-target? Not to limit novelty. The anti-target gate was used as a falsification check: if the workflow could only succeed by falling back into symptomatic cholinergic chemistry, then this would not be a credible root-cause amyloid case study.

Execution workflow

1

Broad selective search

The local discovery workflow enumerated a broad amyloid-oriented BACE1 discovery space under the strict BACE1 + AChE + KCNH2 + CNS ADMET brief.

2

Verubecestat-series refinement

A focused scaffold sweep validated hydroxymethyl-led rescue motifs and established a clean verubecestat-like subseries as a second shortlist lane.

3

Lanabecestat-series refinement

A second focused sweep exposed an underweighted lanabecestat-like dual-hydroxymethyl motif that later became the top global hit.

4

Production rerun and native confirmation

The validated motifs were fed back into scaffold priorities and both the broad and native-only BACE1 spaces were rerun through the same local discovery workflow.

Interpretation: the workflow did not just rank molecules. It identified where the discovery bottleneck moved. By the end of the campaign, the main failure modes were narrow ADMET misses rather than lack of BACE1 signal, which is exactly what a serious triage engine should expose before wet-lab work begins.

How long the published study took

The final public case study is based on the confirmed benchmark-path reruns, not on earlier exploratory passes. Those confirmation runs finished fast enough to function as an active discovery workflow rather than a slow batch report.

11m 57s

Broad confirmation rerun

707 candidates evaluated under the full BACE1 + AChE + KCNH2 + CNS gate.

11m 31s

Native-only confirmation rerun

675 BACE1-native candidates evaluated under the same hard constraints.

23m 28s

Combined full confirmation time

Two full production reruns on the local benchmark path, without GPU refinement.

Typical fragmented workflow

Target hypothesis, scaffold sourcing, potency scoring, counter-screening, and CNS triage are often split across different tools and handoffs.
Strong binders can survive too long before hERG, permeability, or exposure liabilities are made explicit.
Human review time accumulates across hours to days even before assay design starts.

This study's integrated path

The same campaign enforced target potency, anti-target logic, and CNS ADMET together.
Near-miss reasons were visible immediately, so the next medicinal chemistry move was obvious.
The full confirmation package for the published page landed in under twenty-four minutes.

Results overview

The final case-study packet does not depend on a single lucky scaffold. After focused refinement, FluxMateria returned a strict shortlist containing 48 broad-space passes and 34 native-space passes, with lanabecestat-like and verubecestat-like chemotypes both surviving the full gate. That is the core novelty here: one amyloid hypothesis, two scaffold families, both still alive after potency, anti-target, and CNS safety pressure were applied together.

48 / 707

Broad strict passes

ADMET rejections dropped to 1020 after scaffold tuning.

34 / 675

Native-only strict passes

The same top hit survived the tighter native BACE1 search space.

7 / 100

Verubecestat sweep

Hydroxymethyl and CH2F plus CH2OH motifs validated cleanly.

4 / 87

Lanabecestat sweep

Dual-hydroxymethyl chemistry opened the final top-ranked series.

Top final hit

Candidate FM-AD-01, a lanabecestat-like dual-hydroxymethyl analog.
BACE1 11.09, AChE 6.22, hERG 4.85, permeability -5.17, PPB 0.872.

CC(C)Oc1ccc(-c2nc3c(s2)n(-c2cc(F)c(CO)c(CO)c2)c(=O)n3C)cc1

Top shortlist

The shortlist below contains the strongest final strict-pass candidates after the broad rerun. The downloadable JSON packet now includes ranked metrics, series labels, motifs, and SMILES for the top finalists, while near-miss structures and internal workflow details remain private.

Candidate	Series	BACE1	AChE	hERG	Perm.	PPB	Why it stayed
FM-AD-01	Lanabecestat-like Dual-hydroxymethyl	11.09	6.22	4.85	-5.17	0.872	Best overall balance of potency, permeability, and hERG margin.
FM-AD-02	Verubecestat-like Single-hydroxymethyl	11.08	6.26	4.92	-5.33	0.753	Almost tied for first and materially cleaner on PPB.
FM-AD-03	Lanabecestat-like OH + CH2OH	11.05	6.19	4.80	-5.67	0.901	Confirms the new scaffold family is not a one-hit artifact.
FM-AD-04	Lanabecestat-like Dihydroxy fluoro core	11.01	6.16	4.64	-5.49	0.923	Most conservative hERG value in the top tier.
FM-AD-05	Verubecestat-like F + CH2OH	11.01	6.22	4.95	-5.35	0.783	Shows the original series remains competitive after refinement.

FM-AD-01 Lanabecestat-like

CC(C)Oc1ccc(-c2nc3c(s2)n(-c2cc(F)c(CO)c(CO)c2)c(=O)n3C)cc1

Dual-hydroxymethyl lead. Ranked first in both the broad and native-only confirmation reruns.

FM-AD-02 Verubecestat-like

CC(NC(=O)c1ccc2c(c1)CN(C)C(=O)C2)c1nc2cccc(CO)c2s1

Single-hydroxymethyl series. Nearly tied for first while carrying a cleaner PPB profile.

FM-AD-03 Lanabecestat-like

CC(C)Oc1ccc(-c2nc3c(s2)n(-c2cc(F)c(O)c(CO)c2)c(=O)n3C)cc1

Hydroxy plus hydroxymethyl analog. Confirms the top scaffold family is not a single-molecule artifact.

FM-AD-04 Lanabecestat-like

CC(C)Oc1ccc(-c2nc3c(s2)n(-c2cc(O)c(O)c(F)c2)c(=O)n3C)cc1

Dihydroxy fluoro core. The most conservative hERG profile in the top tier.

FM-AD-05 Verubecestat-like

CC(NC(=O)c1ccc2c(c1)CN(C)C(=O)C2)c1nc2ccc(F)c(CO)c2s1

Fluoro plus hydroxymethyl analog. Keeps the second scaffold family strongly competitive after refinement.

What was rejected and why

The value of the workflow is not only what it ranked. It is also what it removed. The highest-ranked failures are mostly narrow ADMET misses, which is the right failure mode for a mature shortlist.

Permeability near miss

A verubecestat-like regioisomer kept BACE1 11.08 and AChE 6.26, but permeability dropped to -6.09, just outside the hard CNS gate of -6.0.

Second permeability miss

A fluoro regioisomer again kept the binding profile but landed at -6.13 permeability. Small positional changes now decide pass versus fail.

hERG-driven rejection

A dimethoxy BACE1 analog held strong potency but moved to hERG 5.17, above the hard cutoff of 5.0, and was removed immediately.

Lanabecestat hERG miss

An amino plus hydroxymethyl lanabecestat analog showed the right amyloid-directional signal, but hERG 5.09 was still too narrow for the strict production screen.

Interpretation

The result is not just that FluxMateria found strong numbers. The more important point is that the campaign behaved like a real discovery workflow: it opened a second chemotype, explained the surviving tradeoffs, and exposed exactly why the best near-misses failed.

Root-cause framing held

The workflow did not collapse back into symptomatic cholinergic chemistry. The AChE anti-target did its job.

Two real chemotypes emerged

Lanabecestat-like and verubecestat-like BACE1-native series both survived the strict screen, which makes the narrative stronger than a one-scaffold story.

The bottleneck moved to real medicinal chemistry

The top failures are now narrow permeability and hERG misses, not absence of BACE1 strength. That is where a serious triage engine should end up.

The lead is robust

The same dual-hydroxymethyl lanabecestat analog ranked first in both the broad and native-only reruns.

Why that matters: this is the real value of high-quality in-silico discovery. FluxMateria already did the discovery, ranking, counter-screening, and prioritization work that normally leaves teams with too many weak candidates and too little clarity. By the time the page ends, the obvious next step is no longer more searching. It is focused experimental validation on the molecules with the highest predicted chance of success.

What happens next

This page should be read as a candidate-discovery case study, not a clinical claim. FluxMateria has already done the in-silico narrowing: it moved the campaign from a broad Alzheimer's search problem to a small, interpretable assay package built around the candidates with the highest predicted probability of surviving potency, selectivity, and early safety pressure.

What This Study Achieved

The remaining step is experimental validation in a lab. That is exactly the point of the study: not to replace wet-lab work, but to make that wet-lab work sharper, cheaper, and more defensible by sending forward a much smaller set of higher-quality candidates.

Done in silico with FluxMateria: BACE1 activity was predicted and prioritized across the top shortlist and the strongest near-miss rescues. Next: confirm experimentally.
Done in silico with FluxMateria: AChE selectivity and hERG safety margins were already counter-screened computationally. Next: confirm experimentally.
Run cell-based amyloid-beta production assays.
Confirm permeability and microsomal stability experimentally.

Important limitation

All values reported here are computational predictions from FluxMateria. This case study supports an amyloid-pathway targeting narrative through BACE1 modulation. It does not prove plaque clearance, disease reversal, clinical efficacy, or suitability for patient care.

Research use notice: This page describes in-silico predictions and preclinical discovery workflows only. 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 use would require experimental validation, additional preclinical work, regulatory review, and where applicable controlled clinical trials.

Full White Paper

Download the complete Alzheimer's amyloid-pathway case study PDF

The full white paper packages the study narrative, mechanism-of-action framing, shortlist interpretation, runtime context, literature positioning, and supporting figures into a single document for sharing with collaborators, partners, or internal review.

Download full white paper (PDF) Direct PDF link

Want the public summary packet?

The public JSON packet includes shortlist metrics, chemotype labels, key findings, and SMILES for the top finalists. Near-miss structures and internal workflow records are retained internally.

Download public JSON Request pilot access

48 strict amyloid-pathway hits from 707 candidates. Two independent BACE1 chemotypes.