SomaNav
Autonomous Scientific Partner

De-risk.
Discover.
Repeat.

Iterative Intelligence

De-risked
Discovery.

SomaNav is an autonomous scientific partner that generates novel, mechanistically grounded hypotheses — and improves with every iteration.

114
Soma Nodes
GLP-1 / AD Session
46
Hypotheses Generated
Across 20 iterations
90%
Top Confidence
Peer-reviewed citations
10/10
Validation Score
PCSK9 session, iteration 3
SomaNav is not a tool. It is a partner. It does not wait to be asked — it identifies what to research next, executes the investigation, critiques its own output, and updates its knowledge model. Every iteration, it becomes more capable. Every session, the graph grows richer.

The bottleneck in drug discovery is not capital. It is the quality of the hypothesis that capital is deployed against.
SomaNav compresses the hypothesis generation phase from months to hours — and improves hypothesis quality with every iteration.

SomaNav — Strategic Self-Planning Session, 47 Iterations
[The Reasoning Loop]

Four phases.
Infinite iterations.

Every session runs the same four-phase cycle — autonomously, without human prompting between iterations. Each phase is executed by a specialised reasoning module. The loop is self-directing: it decides what to investigate, retrieves live evidence, evaluates its own output, and updates its knowledge model.

01
Plan

SomaNav decomposes the research goal into specific sub-hypotheses and identifies which mechanisms, pathways, and entities to investigate in this iteration.

02
Execute

Live literature retrieval, structural biology enrichment, and mechanistic reasoning. Every claim is grounded in peer-reviewed evidence with full citation provenance.

03
Critique

The Critic module evaluates the output against five dimensions: mechanistic coherence, evidence quality, novelty, testability, and commercial relevance.

04
Reflect

The system updates its knowledge model, writes memory entries, and adjusts the next iteration's strategy based on what was learned — and what was contested.


[What SomaNav Produces]

Evidence-backed science,
without the manual work.

A self-improving research intelligence that surfaces novel, testable hypotheses — grounded in peer-reviewed literature, enriched with structural biology, and refined across every session.

01
Ranked Hypotheses

Novel, testable scientific hypotheses with confidence scores, supporting evidence, and full citation provenance from peer-reviewed literature.

"GLP-1R activation upregulates BDNF via cAMP/PKA/CREB — 90% confidence, 8 supporting papers"
02
Soma Knowledge Graph

A persistent, cross-session graph of every entity, relationship, and contested claim — growing richer with every iteration.

114 nodes · 196 edges — GLP-1 / Alzheimer's session, 20 iterations
03
Structural Validation

Key protein nodes enriched with AlphaFold pLDDT confidence scores and PAE heatmaps — structural evidence layered directly into the knowledge graph.

BACE1 pLDDT 90.1 · GLP1R pLDDT 81.5 · BDNF pLDDT 71.6
[Case Study 01]

GLP-1 receptor agonists
for Alzheimer's.

20 iterations. 114 Soma nodes. 46 hypotheses. The system mapped the full intracellular cascade, identified three high-confidence therapeutic mechanisms, and flagged the amyloid hypothesis as contested — without human direction at any step.

[Mechanistic Pathway]
Semaglutide
GLP-1R
cAMP / PKA
BDNF / TrkB
GSK-3β → Tau
BACE1 / APP
⚠ Amyloid-β — Contested. Tau and neuroinflammation pathways assigned higher causal weight.
90%

GLP-1R activation in hippocampal neurons upregulates BDNF expression via cAMP/PKA/CREB signalling, promoting neuronal survival and synaptic plasticity.

PubMed · 8 supporting papers · AlphaFold pLDDT 81.5
90%

Semaglutide-mediated GLP-1R activation inhibits GSK-3β, reducing tau hyperphosphorylation at Ser396/Thr231 and slowing neurofibrillary tangle formation.

PubMed · 6 supporting papers · ELAD trial NCT04777396
85%

GLP-1R agonism suppresses neuroinflammatory signalling through TNF-α and IL-6 pathway inhibition, reducing microglial activation in high amyloid-β burden regions.

PubMed · 5 supporting papers · Amyloid pathway contested
[Findings vs. Published Literature]

GLP-1 / Alzheimer's —
how SomaNav compares.

SomaNav's session independently converged on the GSK-3β / Tau phosphorylation axis and the BDNF / TrkB neuroprotection pathway. Below is how those findings align with the published clinical and preclinical record.

◆ SomaNav — Session 240001 · 20 Iterations

GSK-3β hyperactivation drives Tau hyperphosphorylation at Ser396/Thr231. GLP-1R agonism restores AKT → GSK-3β inhibition, reducing neurofibrillary tangle formation. BDNF / TrkB upregulation provides parallel neuroprotection via CREB-mediated transcription.

Mechanism identified · Iteration 4 · Confidence 81%
Pathway: GLP-1R → cAMP → PKA → AKT → GSK-3β(Ser9) → Tau
81%SomaNav confidence score
○ Published Literature

Liraglutide ameliorates neuronal insulin resistance and reduces Tau hyperphosphorylation via AKT–GSK-3β signalling. Phase IIb trial (ELAD, 2024) showed liraglutide slowed cognitive decline in early Alzheimer's patients. Semaglutide associated with 40–70% reduced risk of first-time Alzheimer's diagnosis in T2DM cohorts.

Jantrapirom et al., PMC7084306 · Liraglutide / GSK-3β / Tau, 2020
ELAD Phase IIb Trial · AAIC 2024 · Novo Nordisk
Nørgaard et al., Alzheimer's & Dementia, Oct 2024
40–70%Reduced AD risk (semaglutide vs. other antidiabetics)

SomaNav's mechanistic pathway was generated autonomously without access to these papers. The convergence with published clinical evidence is a measure of mechanistic coherence, not a claim of clinical efficacy. All SomaNav outputs are research hypotheses and require independent experimental validation.

View Full Benchmark Report
[Case Study 02]

PCSK9 inhibitors &
ANGPTL3 co-inhibition.

15 iterations. The system independently identified ANGPTL3 as an upstream regulator of LPL activity and proposed combination therapy with Inclisiran and Evinacumab — a synergistic mechanism supported by ELIPSE trial data.

[Validation Score — Iteration 3]
10 / 10
"Pathway mapping is complete and mechanistically coherent. ANGPTL3 intersection is well-supported by ELIPSE trial data."
[Entities Discovered]
PCSK9ANGPTL3InclisiranEvinacumabLDLRLPL
[Top Hypothesis]

Dual PCSK9/ANGPTL3 inhibition achieves synergistic LDL-C reduction exceeding 70% in statin-refractory patients through complementary mechanisms of action.

88% confidence · ELIPSE trial · 7 supporting papers
[Findings vs. Published Literature]

PCSK9 / ANGPTL3 —
how SomaNav compares.

SomaNav independently identified ANGPTL3 as an upstream LPL regulator and proposed dual inhibition with Inclisiran and Evinacumab. Here is how that maps to the clinical evidence base.

◆ SomaNav — Session 270001 · 15 Iterations

ANGPTL3 inhibits LPL activity, raising TG and LDL-C. Dual blockade of PCSK9 (via Inclisiran siRNA) and ANGPTL3 (via Evinacumab) achieves synergistic LDL-C reduction exceeding 70% in statin-refractory patients through complementary mechanisms of action.

Mechanism identified · Iteration 3 · Confidence 88%
Pathway: ANGPTL3 → LPL inhibition → TG/LDL-C elevation → ASCVD risk
>70%Predicted LDL-C reduction (dual blockade)
○ Published Literature

ELIPSE HoFH trial: Evinacumab reduced LDL-C by ~49% in homozygous FH patients on maximally tolerated statins. ANGPTL3 inhibitors achieved greater TC reductions than PCSK9 inhibitors alone (−49.9% vs −21.2%).

ELIPSE HoFH Trial · Raal et al., NEJM 2020
Bytyçi et al., meta-analysis, PubMed 41501311, 2026
Ray et al., JACC 2025 · Solbinsiran / ANGPTL3
−49.9%TC reduction (ANGPTL3 inhibitors, meta-analysis 2026)

SomaNav's dual-inhibition hypothesis was generated before these meta-analyses were published. The mechanistic logic is consistent with the ELIPSE trial data. All outputs are research hypotheses requiring independent experimental validation.

View Full Benchmark Report
[Soma Compass]

The system that chooses
what to research next.

Every research tool tells you how to investigate a question. SomaNav is the first to tell you which question to investigate. After each session, Soma Compass analyses the entire knowledge graph, identifies the three highest-value unexplored territories, and proposes the next sessions — each with a mechanistic rationale, a pre-filled goal, and a one-click approval.

01
Analyse the Soma graph

After each session, SomaNav reads every node, edge, confidence score, and contested relationship in the knowledge graph. It measures density, uncertainty, and commercial relevance across the entire network.

02
Score every gap

A four-dimension priority function scores each unexplored territory: knowledge gap density, confidence deficit, commercial relevance, and template coverage. The top three candidates are selected.

03
Generate proposals

For each candidate, a mechanistic rationale is generated — identifying the specific Soma nodes driving the gap and writing a pre-filled goal statement ready to launch immediately.

04
You approve — one click

Three proposals. You read the rationale, review the gap nodes, and click Approve. A session is created instantly with the pre-filled goal. The system executes. You direct.

[Priority Function — 4 Dimensions]
35%
Knowledge Gap Score

Inverse edge count per node. Sparse areas of the Soma graph score highest. Contested nodes receive a 1.5× multiplier — unresolved contradictions are high-value targets.

30%
Confidence Deficit

High-weight nodes with low confidence represent important concepts that are not yet well-established. Score: (1 − confidence) × weight / 10.

20%
Commercial Relevance

Domain-weighted relevance to your active research programmes. Cross-domain proposals receive a 1.5× novelty multiplier for bridging separate fields.

15%
Template Coverage

Unused session templates score 0.8. Templates run once score 0.5. Templates run three or more times score 0.2. Ensures systematic coverage of the knowledge space.

value_score = (gap × 0.35) + (deficit × 0.30) + (relevance × 0.20) + (coverage × 0.15)
Drug Discovery · Structural Biology

Molecular intelligence,
visualised in 3D.

SomaNav's drug discovery pipeline surfaces validated targets and hit compounds from live bioactivity databases. Every molecule is rendered from crystallographic data — the same structures used by medicinal chemists at leading pharma companies.

SemaglutideIbuprofenAspirinCaffeineCollagen Peptide
SourceRCSB Protein Data Bank
Render3Dmol.js · WebGL
PipelineTarget ID → Lead Optimisation
PDB: 7KI0 · RCSB
SemaglutideC₁₈₇H₂₉₁N₄₅O₅₉
GLP-1 Receptor Agonist
↳ GLP-1R / Glucagon-like peptide-1 receptor
Type 2 Diabetes
Obesity
Cardiovascular Risk
[Get Started]

Your autonomous
scientific partner.

Start your first session in minutes.
No qualification process. No enterprise contract.

Initiate Iteration →View Investor Demo