Evaluate indication expansion without hiding the weak dimension

An indication ranking often looks rigorous because every row has a number. The false precision usually began earlier, when different patient populations, evidence depths and development paths were forced into the same template.

Resolve the asset and patient segment

“Lung cancer” is not an indication strategy. The relevant unit may include histology, biomarker, treatment line, prior therapy and geography. ARiDA resolves that decision unit before scoring and records the positioning mode: broad population, biomarker-led entry, niche beachhead or lifecycle expansion.

The system then selects weights that match the decision archetype. An early biology screen should not use the same emphasis as a late-stage lifecycle decision. Management can override weights, but the override remains visible and sensitivity analysis tests whether the ranking survives reasonable changes.

Build each dimension from evidence

The science side can include target validation, pathway enrichment, druggability, assay quality, ADMET readiness, translational support and safety. This is not academic decoration. Recent work continues to find a materially higher probability of clinical success for genetically supported mechanisms, while showing that the effect varies by context (Minikel et al.).

Clinical feasibility needs the proposed population, endpoints, precedent, recruitment burden and competitive trial environment. Commercial attractiveness needs a defined patient funnel, treatment duration, price and plausible share. FDA’s TPP framework helps keep desired claims tied to the evidence program required to support them (FDA).

ARiDA assembles this evidence into frozen, source-bound records before the deterministic kernel runs. The ranking engine cannot create facts. If positive cost and revenue assumptions are not both source-backed or explicitly owned by an operator, it emits an economics gap instead of inventing a value.

Keep gates outside the average

A serious safety liability, missing causal hypothesis or infeasible trial should not disappear inside a weighted average. ARiDA evaluates decision gates separately. Unknown gate evidence remains unresolved. A hard failure can remove an indication from the investable set even when its commercial score is high.

For viable indications, the studio can calculate per-indication rNPV and compare value with biology fit, evidence confidence and development constraints. Clinical benchmark probabilities remain cohort-specific; the code does not blend unlike reference classes into a comforting universal rate. That restraint matters because success varies sharply by disease, phase and trial design (Wong, Siah and Lo).

Sequence, do not merely rank

The best first indication may create proof of mechanism, regulatory leverage or a commercial foothold for the second. ARiDA’s sequence optimizer can account for dependencies, resource limits and the value of learning. The result is a development sequence with stated reasons, binding constraints and experiments that could move an indication.

A useful ranking tells the team where to go. A serious indication strategy also tells it what must become true.