SomaNav was not built to summarise research. It was built to do what no human researcher can: read everything, forget nothing, challenge every assumption, and compound its own understanding across every session it runs — indefinitely.
The question that started everything: why does so much scientific knowledge never become anything?
Aves Wencel did not come to biomedical research through a laboratory. He came through a different kind of obsession — the kind that starts with a question you cannot stop asking, and that refuses to be answered by any single discipline.
The question was this: why does so much scientific knowledge never become anything? Decades of published research. Thousands of promising compounds. Entire fields of mechanistic understanding — and yet the gap between discovery and treatment remains vast, and growing. The problem is not a lack of data. It is a lack of synthesis.
"The bottleneck in almost every domain is not intelligence. It is the architecture of how knowledge compounds — or fails to."
The tools built to help researchers — literature databases, AI summarisers, chatbots with search plugins — all share the same fundamental flaw. They retrieve. They do not reason. They respond to prompts. They do not pursue hypotheses. They produce outputs. They do not accumulate understanding.
SomaNav began as a personal experiment: could a machine be built that does not just retrieve information, but genuinely accumulates understanding? One that critiques its own reasoning, identifies what it got wrong, and uses that failure as fuel for the next iteration? The answer, after hundreds of iterations across drug repurposing, longevity therapeutics, and molecular pathway analysis, is yes.
The conviction: the most important breakthroughs in medicine will not come from a single genius in a single lab — they will come from a machine that never stops thinking.
The global biomedical literature grows by more than one million papers per year. Drug discovery takes an average of 12–15 years and costs over $2 billion per approved compound. The failure rate in clinical trials exceeds 90%. Most of that failure is not because the science is wrong — it is because the right connections between existing knowledge were never made. The synthesis gap is not a research problem. It is an infrastructure problem. SomaNav is the infrastructure.
SomaNav operates on a proprietary iterative learning architecture that continuously refines hypotheses, expands its knowledge base, and evolves its capabilities through a cycle of planning, execution, and critical self-reflection. Each iteration builds on the last. The system scores its own output, identifies what failed, and redesigns its approach before the next cycle begins. Improvement is the default state.
Every hypothesis, finding, and validated connection is stored in the Soma knowledge graph — a permanent, growing map of what SomaNav has learned. Nodes represent concepts, entities, and mechanisms. Edges represent evidence-weighted relationships. The graph never resets. It compounds. A session run today builds on every session run before it, across every domain SomaNav has explored.
Every claim SomaNav makes is cross-referenced against live scientific databases: PubMed, Semantic Scholar, OpenAlex, ChEMBL, UniProt, Open Targets, and ClinicalTrials.gov. Citations are attached to every output. Nothing is asserted without evidence. The system is designed to be wrong in a useful way — with a traceable trail of reasoning that can be challenged and corrected.
SomaNav owns the full pre-molecular drug discovery pipeline: Target Identification → Target Validation → Hit Discovery → Lead Optimisation. The output is a structured pre-clinical dossier, ready for handoff to molecular design partners. Every stage is grounded in live data and scored for confidence.
Disease → ranked target list with mechanistic rationale, genetic evidence, and pathway context.
Druggability scoring, AlphaFold structure, tissue expression, safety flags, confidence grade.
ChEMBL bioactivity compounds, Lipinski Rule-of-5, IC50/Ki/Kd, selectivity profile.
AI SAR analysis, structural modifications, ADMET prediction, binding affinity estimation.
Structured dossier exported to CRO or molecular design partner.
Phase I–III trials. SomaNav continues monitoring literature for emerging evidence.
| Capability | SomaNav Autonomous loop | Isomorphic Labs Molecular design | Perplexity / ChatGPT General AI search | Elicit / Consensus Literature AI |
|---|---|---|---|---|
| Autonomous multi-step reasoning | ✓ | — | — | — |
| Persistent knowledge graph (never resets) | ✓ | — | — | — |
| Self-critique and score-driven improvement | ✓ | — | — | — |
| Target ID → Lead Optimisation pipeline | ✓ | Partial | — | — |
| Protein structure + molecular design | AlphaFold viewer | ✓ (proprietary) | — | — |
| Live citation grounding (PubMed, ChEMBL, etc.) | ✓ | — | Partial | ✓ |
| FMCG / nutraceutical domain support | ✓ | — | Generic only | — |
| Self-service (no enterprise contract) | ✓ | Enterprise only | ✓ | ✓ |
| Pre-clinical handoff dossier export | ✓ | — | — | — |
The global biomedical literature grows by more than one million papers per year. No human researcher can synthesise it. No single institution can act on it. SomaNav exists to change that.
Every hypothesis SomaNav generates is cross-referenced against PubMed, OpenAlex, ChEMBL, Reactome, and ClinicalTrials.gov. Nothing is published without a citation trail.
The Critique and Reflect modules exist to find what went wrong. A low score is not a failure — it is data. Critical failures trigger targeted root-cause analysis and recovery protocols.
The full benchmark report is public. The scoring methodology is documented. The system's self-assessment is compared against external evidence on every iteration.
Retrieval is a commodity. Compounding is rare. SomaNav is not optimised to find information faster — it is optimised to build understanding deeper. The Soma graph grows with every session.
The most important discoveries happen at the intersection of disciplines. SomaNav works across drug discovery, nutraceuticals, longevity science, and FMCG innovation.
SomaNav does not wait for a human to improve it. The Evolution system tracks performance across sessions, unlocks new capabilities as the system matures, and continuously recalibrates.
Nine-step autonomous reasoning loop with live literature grounding. Full Target ID → Lead Optimisation pipeline. Soma knowledge graph. Session templates for biotech, drug repurposing, and FMCG. Pre-clinical handoff dossier export.
3Dmol.js native renderer — photorealistic protein surfaces with electrostatic colouring, binding site highlighting, and animated rotation. Advanced protein-ligand co-folding integration for binding affinity prediction.
Integration of cutting-edge algorithms for AI-driven de novo molecule generation optimised for validated targets. This closes the gap between SomaNav's lead optimisation output and Isomorphic-style molecular design.
Parallel autonomous agents running independent hypotheses on the same target, with a meta-critic synthesising findings across agents. Designed for organisations running multiple research programmes simultaneously.
SomaNav is currently in private beta. Request access to run your own research session, or explore the public benchmark report to see what the system has already produced.