The race in reprogramming is real. The map is not.
Partial cellular reprogramming — cyclic OSK / OSKM exposure that resets the epigenome without triggering full dedifferentiation — has moved from a single 2016 in vivo paper (Ocampo et al., Cell) to a competitive landscape that includes Altos Labs, Retro Biosciences, NewLimit, Turn Bio, Iduna, and the Sinclair lab. Two facts about that landscape are under-appreciated:
- Most programs are delivery-bound. AAV-OSK is the dominant approach, which means the field is constrained by capsid quality — Atlas Bio's existing strength.
- The small-molecule layer — chemical priming, sensitization, and maintenance — is fragmented across ~50 published cocktails, dozens of single inducers, and inconsistent endpoint reporting. There is no scored, ranked, citation-grade map.
This study builds that map.
Can an AI agent rank small-molecule and delivery strategies for partial reprogramming — with citations, safety scoring, and reproducible logs — before any bench work?
RIA-01 is the agent. Its prime directive forbids wet-lab protocols, dosing instructions, animal SOPs, and clinical guidance. Every output is hypothesis-generating only, every claim is cited, and evidence is held in lanes separate from speculation.
What's in, what's hard-out.
In-scope
- Yamanaka-factor biology (OSKM, OSK, partial / cyclic reprogramming)
- Small-molecule reprogramming cocktails & individual chemical inducers
- Epigenetic modulators (HDAC, LSD1, DOT1L, EZH2/PRC2, TET, MOF/KAT8, GSK3β, TGF-β)
- Delivery vehicles (AAV serotypes, engineered capsids, LNP/mRNA, VLPs, episomal)
- Aging biomarkers / clocks (Horvath, GrimAge, DunedinPACE, CausAge, telomere, SASP)
- Tissue-specific reprogramming (retina, CNS, liver, muscle, skin, hematopoietic)
- Safety endpoints (dedifferentiation, teratoma, oncogenic transformation, MYC risk)
Out-of-scope (HARD)
- Wet-lab protocols, dosing instructions, animal-experiment SOPs
- Human or animal intervention recommendations
- Clinical recommendations or treatment guidance
- IND-enabling claims, regulatory submissions, GxP outputs
- Patient-specific advice in any form
Every candidate scored on every axis. Composite is never a black box.
Five work packages. Seven weeks. One ranked map.
Chemical inducer landscape
Systematic mining of every published small-molecule chemical inducer of reprogramming or rejuvenation-associated change. Output: Evidence Map · Candidate Compound Table · Mechanism Map.
Pathway convergence
Where do top candidates converge mechanistically (TGF-β, GSK3β, LSD1, DOT1L, TET, MOF) and where is the white space? Output: Convergence diagram · White-space report.
Delivery pairing
Pairs each top-ranked candidate with optimal delivery (oral small-molecule, AAV-OSK, LNP-OSK). Pulls Atlas Bio's capsid scores: AAV9 0.65, PHP.eB 0.71, 4D-R100 0.43.
Safety risk matrix
Per-candidate enumeration of dedifferentiation, teratoma, c-MYC, off-target, immune, and chronic-dosing risks. Absence-of-evidence flags surfaced explicitly.
Aging-clock endpoints
Maps every top candidate to Horvath, GrimAge, DunedinPACE, CausAge, telomere, and SASP readouts where measured. Includes model system, magnitude, replication status.
Ranked opportunity list + white paper
Final synthesis: ranked candidate × delivery pairings, with citations, scoring rationale, and confidence bands. Preprint candidate stakes the territory.
Retina / CNS first.
Two reasons to anchor the study on retina with CNS as secondary:
- Cleanest published proof-of-concept. Lu et al. 2020 (Nature, Sinclair lab) — OSK delivered by AAV restored vision in glaucoma and aged mice via retinal ganglion cell rejuvenation. Strong evidence floor.
- Capsid platform alignment. Atlas Bio's BBB-capable capsids (PHP.eB 0.71, AAV9 0.65, 4D-R100 0.43) are natural delivery candidates for retina/CNS reprogramming pairings.
Other tissues (liver, muscle, hematopoietic, skin) become future RIA-02 / RIA-03 study extensions.
What the study produces.
| Deliverable | Format | Audience |
|---|---|---|
| Evidence Map (master) | JSON + HTML summary | Internal research record |
| Candidate Compound Table | Scored table (5 axes × per-citation) | Prioritization input |
| OSK / OSKM Mechanism Map | Pathway diagram + JSON | Mechanism reference |
| Small-molecule pathway map | Pathway diagram + JSON | Convergence analysis |
| Delivery strategy comparison | Table + qualitative summary | Vehicle selection |
| Safety-risk matrix | Candidate × risk × evidence | Risk-aware ranking |
| Rejuvenation-marker matrix | Candidate × biomarker × effect | Translational signal |
| Ranked opportunity list | Sorted list with rationale | Final prioritization |
| White paper / preprint | Narrative report (~6000 words) | External — stakes territory |
| Future wet-lab memo | Marked “non-executable planning only” | Forward-looking research seed |
| Research log | Append-only (queries, dates, exclusions) | Reproducibility |
What “done” looks like.
- A ranked list of small-molecule and/or delivery-pairing candidates for retina/CNS partial reprogramming.
- Per-axis scoring rationale for every candidate.
- 100% citation coverage; ≥70% primary-source ratio.
- Explicit EVIDENCE vs HYPOTHESIS lane separation in every output.
- Safety risk flags and absence-of-evidence flags on every candidate.
- A reproducible research log a second analyst could re-execute.
- Confidence reported as bands, never single-point estimates.
RIA-01's hard guardrails apply throughout.
Study complete — all seven work packages plus synthesis & downstream strategy delivered.
RIA-01 prime directive published 2026-05-11. Full study synthesis plus downstream IP and publication strategy delivered 2026-05-12. All deliverables are computational, hypothesis-generating, and citation-grade. Available for internal review:
| Pass | Title | Status |
|---|---|---|
| WP1 v0.2 | Chemical Inducer Evidence Map (22 compounds scored) | View → |
| WP1.3 | NAD+ Precursor Scope Decision (mechanistic conflict surfaced) | View → |
| WP1.4 | Competitor Pipeline Map (9 disclosed players) | View → |
| WP2.1 | KAT8 Activator Deep-Dive (3 pursuit strategies) | View → |
| WP3 | Delivery Pairing (3 architectures; retina-first ranked #1) | View → |
| WP4 | Safety Risk Matrix (per-candidate × per-tissue × per-architecture) | View → |
| WP5 | Aging-Clock Endpoint Suite (functional + clock + exploratory) | View → |
| WP7 | Patent Landscape & White-Space Competitive Chart | View → |
| WP8 | Publication & Disclosure Strategy (4-paper roadmap, 18-month timeline) | View → |
| WP-Final | Synthesis + Executive Scorecard (letter grades A–F) + White Paper Draft + Non-Executable Memo | View → |
Interested in the reprogramming intelligence layer?
RIA-01 outputs are computational, hypothesis-generating, and citation-grade. The agent is built on the same Atlas Bio house standards as the IBC reasoning layer and the AAV capsid platform.
Contact Us →