Cellular intervention design framework

A concise reference to the ideal intervention design template, the C0 to C3 explicitness classes, and the four reprogramming setting constraint regimes used to classify indexed methods. Full definitions, class equations, and borderline cases are in the companion review and S1 Text.

Ideal intervention design template

Indexed computational methods for cell reprogramming share a biological aim: nominate perturbations that move a source cell state toward a desired target. The equation below is an idealized template for that inverse problem. It is not a claim that every indexed method already solves it explicitly; classes C0 to C3 record how completely each method instantiates the template.

Hover or focus a block to read its role in the ideal template. The equation states a canonical objective, not the optimization procedure any particular indexed method employs.

Select a block in the schematic.

u^{⋆} = ar g u \in U min [D (F (P_{S}, u), P_{T}) + λ R (u)]

Idealized intervention design template (Section 2 of the companion review). Classes C0, C1, C2, and C3 differ in how much of this objective each indexed method instantiates.

Inverse intervention explicitness (C0 to C3)

Indexed methods are classified by how completely they instantiate the intervention design template above. Explicitness refers to which mathematical object is present in the formulation, not biological accuracy, causal identifiability, or held out performance. Not every method that outputs actionable perturbations is a formal inverse design method; higher class does not mean a better method.

C0 Target-feature and screen-based discovery 7 methods C1 Network informed intervention prioritization 10 methods C2 Proxy inverse design 10 methods C3 Explicit model based inverse intervention 32 methods

Class C0 methods can be valuable candidate discovery tools without being formal inverse design methods. Explicitness increases from C0 to C3 as more objects from the shared template enter the formulation. Select a class for its equation, definition, and indexed methods. Counts are single-class publications (7, 10, 10, and 32). OCSANA+ is assigned to C1 and C3 and is not included in those totals.

Boundary conditions

Class records how a method selects the intervention. Representation, controllability, and evaluability decide what that selection can claim. A higher class does not improve those three conditions.

Representation

Representation is which cell-state properties the method encodes, and which of those the mismatch uses to compare a response with the target. A selector that sees only those coordinates cannot tell apart two sources that share them but need different interventions.

Controllability

Controllability is what an admissible intervention can achieve from the source, at a stated cost, including timing, intervention type, and context. A score can name a regulator without establishing that a feasible intervention reaches the target.

Evaluability

Evaluability is which observations support the claimed outcome. A signature match, a network score, and a model prediction each support a different claim. Evidence has to match the outcome the method actually selects.

Four reprogramming settings

Pluripotency induction, directed differentiation, transdifferentiation, and phenotype reprogramming are not four unrelated computational questions. They are four constraint regimes on the same intervention design syntax: each setting fixes how source state, target state, admissible interventions, and feasibility penalties enter the template. Select a setting for its definition and indexed methods.