AI-Designed Multi-Pathway Cellular Repair — Research Program v0.1

This research program asks a single question: can a computationally designed, multi-pathway intervention restore function in a stressed cell further than the best single-pathway intervention, without pushing that cell toward proliferation or loss of identity? The program is scoped deliberately narrowly. It targets a cell state rather than a molecule, uses aged human retinal pigment epithelium (RPE) as the first system, and takes phagocytic index — photoreceptor outer-segment particles internalized per cell — as its primary endpoint, measured seven days after compound washout so that durability is intrinsic to the primary test rather than a separate hurdle. Three claims are set out and each is separately testable: that functional decline in a defined cell type is driven by several pathways at once; that coordinated input across those pathways reaches a cell state no single input reaches; and that a closed design-test-learn loop converges on useful combinations faster than human-designed screening. The second is identified as the load-bearing claim, and the conditions under which it would be declared unsupported are fixed in advance. The document surveys six candidate pathway axes in aged RPE against published effect sizes, works through why the obvious single-molecule format cannot reach intracellular targets, and specifies a six-arm design whose comparators include both the best single agent at its own optimal dose and a Bliss-predicted additive combination. Identity guardrails (RPE65, BEST1, MITF retention; no Ki67 or EdU increase; no EMT markers) are treated as exclusions rather than covariates, since de-differentiated proliferating RPE is itself a known pathological state. An appended pre-registration fixes hypotheses, cell models, primary and secondary outcomes, safety exclusions, arms, sample size, analysis plan and stopping rules before any data is collected. It states the design's declared limitations, including that perturbation priors are drawn largely from proliferating cancer lines and may transfer poorly to post-mitotic RPE, and that a positive result would be one cell type performing one function in vitro — not rejuvenation, not a therapy, and not generalizable to other tissues without separate work. No experimental data is reported. This is a program specification and pre-registration, published to establish a dated, citable record prior to experimental work.