A ranked target list is a set of hypotheses ordered by evidence, not a confirmed result. The distance between a high rank and a validated target is the distance between association and causation, and only experiments close it.
What experimental validation typically establishes
- The target is expressed in the tissue and cell type where the disease occurs
- Modulating the target — blocking or activating it — produces the expected change in disease-relevant cellular behavior
- The effect holds in a disease model, not only in an artificial assay system
- The result is reproducible and not an artifact of one experimental setup
Why computational evidence cannot substitute for this step
Genetics, literature, and omics all describe what has already been observed in populations, published experiments, or collected samples. None of them directly tests whether intervening on this protein in a living system will change the disease. That intervention test is exactly what validation supplies, and it is why targets are confirmed experimentally before a program commits to years of molecule design and optimization.
AI's contribution at this stage is not producing correct targets but narrowing the field of plausible ones. A ranking that reliably places the right target in the top handful is valuable even though every candidate in that handful still requires experimental confirmation.