�鶹��ý

Pun et al.’s recent review entitled “Target identification and assessment in the era of AI” in  maps how artificial intelligence (AI) is widening target identification through multi-omics integration, knowledge graphs, and foundation models. The authors place patentability, commercial tractability, and competitor analysis alongside druggability and safety within target assessment itself, before experimental validation. This framework implies a forming tension in AI-driven discovery: faster hypothesis generation makes it harder to decide when to commit to a specific invention. 

AI is rapidly increasing the number of plausible targets and drug candidates, often outpacing experimental validation. Integration of multi-omics data, knowledge graphs, and machine learning enables systematic generation and prioritization at a scale not previously possible. At the same time, generative methods produce large sets of candidates that still need to be filtered through validation, safety, and translational constraints. As a result, the bottleneck is shifting from finding viable options to selecting those that can be validated, differentiated, and advanced with confidence.

AI is compressing discovery timelines, but not decision-making. In that environment, the temptation is to keep the hypothesis fluid — to let the model generate one more iteration — rather than commit. The frameworks and overview of AI approaches provided by Pun et al. will help create a system for selecting targets and drug candidates to develop with greater confidence. By placing patentability alongside scientific and translational criteria, Pun et al.’s frameworks effectively bring IP into target selection rather than leaving it to downstream considerations.

In practice, this can change decisions and help commit to a target. Faced with two biologically plausible targets, teams may favor the one where differentiation can be demonstrated experimentally and translated into defensible patent protection, even if the alternative has stronger initial scientific support but sits in a more crowded landscape.

As AI expands what is possible, the constraint is shifting toward what can be validated, differentiated, and owned.

“The strategic selection of targets often involves a delicate trade-off
between novelty and confidence in the target’s role in disease.
High-confidence targets are supported by more scientific evidence,
offering a clearer translational path to the clinic and reducing the
risk involved in drug development. However, novel targets present
opportunities for breakthrough therapies, especially for diseases with
unmet medical needs.”