Portfolio · Project 22

Alexander CHERNOV

🤖 Link-Think-Act · Principal Engineer · AI & Data Infrastructure · Agentic Datasets · Semantic Gateways · Agentic AI & Scientific Data Systems · IEEE Member · Research & Technical Speaker

← the whole portfolio
22

Gene agents: clinical evidence decided under each agent’s own policy, with a gated quantum step

2026 – present · Independent work, unaffiliated with any employer.

Three agents read the same clinical evidence, and each decides by a policy file it holds and cites; the one quantum call passes a gateway.

An Agent Scope demo on real CIViC evidence for three genes. No controller assigns work, every agent moves only along its state machine, and TP53’s single capability, a QAOA contig assembly on a simulator, is reached through a gateway that keeps its own allow-list.

The gene-agents demo as recorded: about 100 seconds from the running system, every caption quoting the record it describes. Evidence reaches the three agents in the order CIViC's curators reviewed it, each agent decides under its own policy file, and one capability request goes through the gateway to the quantum step. The file is the one archived on Zenodo.

Three agents work on real clinical-evidence records from CIViC, one each for EGFR, TP53 and BRCA1. Items arrive in the order CIViC’s curators reviewed them, every item goes to every agent, and each agent decides for itself whether an item concerns it. An agent is Idle, Reasoning, Publishing or Cooldown, and every transition is a record with the state before and after, so the color on screen is the state the agent was in.

Each agent loads a versioned policy file once and cites its id, version and sha256 on every decision. At the default evidence floor, 140 of 269 EGFR items are refused for therapy claims, because the finding stands but using it to say a treatment will work does not.

TP53 may request one capability: de novo assembly of a short contig as QAOA, following the published QuASeR method, with three reads and nine qubits on PennyLane’s default.qubit simulator. The agents hold no network code, so the call can only go to the gateway, which refuses a backend that is not on its list, serves an exact repeat from cache, or runs the call. The run returns GCGCA, checked against the brute-force optimum and reported with its margin of 0.0005 over the runner-up, and a request for ibm_hardware is refused before anything is submitted.

The trace view is the published @agent-scope-ca/graph-canvas component, available on npm, and the assembly code comes from the public quantum-biology demos.

References

  1. Sarkar, Al-Ars and Bertels (2021). QuASeR: Quantum Accelerated de novo DNA sequence reconstruction. PLOS ONE 16(4), e0249850. doi:10.1371/journal.pone.0249850
  2. Griffith et al. (2017). CIViC is a community knowledgebase for expert crowdsourcing the clinical interpretation of variants in cancer. Nature Genetics 49(2), 170–174. doi:10.1038/ng.3774
  3. Bergholm et al. (2018). PennyLane: Automatic differentiation of hybrid quantum-classical computations. arXiv preprint. arXiv:1811.04968