# Axiom Science 4.5

Axiom Science 4.5 is an AI research workbench: describe tasks in natural language—find target evidence, compute protein features, design variant sequences, search literature—and it picks skills and data sources, runs multi-step work in a sandbox, and grounds every conclusion in real evidence. 32 built-in skills and 96 tools spanning structure and protein design, genomics and clinical interpretation, transcriptomics and single cell, pharmacology and regulation, trials and cohorts, and literature review; ESM2, ProteinMPNN-family, scVI, and more run locally offline. 30+ authoritative sources direct-connected: AlphaFold DB, UniProt, PDB, ClinVar, ChEMBL, openFDA, GTEx, Open Targets, bioRxiv, and more. Conclusions you can check: numbers, IDs, DOIs, and tool outputs compared character-by-character—about three-quarters verified locally; another model reviews and deletes unsupported sentences with reasons. Visible sandbox: skills are containers not scripts—non-root, read-only root FS, network off by default, per-resource limits; code, egress, and GPU need approval. Public tenants isolated; local edition keeps data on your machine. Extensible via SKILL.md, MCP connectors, expert sub-agents, and long-term memory. For structural biology, genomics, drug discovery, and clinical research.

Axiom Science 4.5 is an AI research workbench: describe tasks in natural language—find target evidence, compute protein features, design variant sequences, search literature—and it picks skills and data sources, runs multi-step work in a sandbox, and grounds every conclusion in real evidence. 32 built-in skills and 96 tools spanning structure and protein design, genomics and clinical interpretation, transcriptomics and single cell, pharmacology and regulation, trials and cohorts, and literature review; ESM2, ProteinMPNN-family, scVI, and more run locally offline. 30+ authoritative sources direct-connected: AlphaFold DB, UniProt, PDB, ClinVar, ChEMBL, openFDA, GTEx, Open Targets, bioRxiv, and more. Conclusions you can check: numbers, IDs, DOIs, and tool outputs compared character-by-character—about three-quarters verified locally; another model reviews and deletes unsupported sentences with reasons. Visible sandbox: skills are containers not scripts—non-root, read-only root FS, network off by default, per-resource limits; code, egress, and GPU need approval. Public tenants isolated; local edition keeps data on your machine. Extensible via SKILL.md, MCP connectors, expert sub-agents, and long-term memory. For structural biology, genomics, drug discovery, and clinical research.

Canonical: https://uniquebase.ai/products/axiom-science-4-5
Stable identifier: UB-D7019
Snapshot captured: 2026-11-04T00:48:01.362Z
English updated: 2026-11-04T01:06:54.938Z

## Connected profiles

- [Axiom3](/companies/axiom3)

## Sources

- No source links are available for this profile.