New Ways To Find Drugs

Iterative model–experiment cycles can amplify existing model biases and overfit to the proprietary priors they use, making targets look validated within the loop yet fail to generalize outside it—a risk explicitly noted for lab‑in‑the‑loop workflows that lean on domain atlases like Osteomics as training data osteomics.co.uk pubmed.ncbi.nlm.nih.gov.

Non‑public atlases impede independent scrutiny and replication, increasing the chance that dataset‑specific artifacts drive target nomination and that nominated targets lack causal relevance across broader patient populations, limiting confidence in translation osteomics.co.uk pubmed.ncbi.nlm.nih.gov.

Single‑cell associations are inherently correlational and vulnerable to confounding from inflammation, cell‑composition shifts, technical batch effects, and sampling bias; the need for orthogonal functional validation and causal‑inference methods before advancing targets signals a high evidentiary bar in bone/cartilage programs osteomics.co.uk pubmed.ncbi.nlm.nih.gov.