From Silicon to Carbon: Experimental Validation of Generative AI-Derived CNS Drug Candidates — A Systematic Review (2021–2026)
Year: 2026
Type: preprint
Study design: Systematic Review
Abstract: Background Generative artificial intelligence has gained traction as an approach for designing novel CNS-relevant drug candidates, yet the extent to which computationally generated molecules undergo experimental validation remains poorly characterised. Methods A systematic review was conducted in accordance with PRISMA 2020 guidelines using a prospectively registered protocol (OSF: https://doi.org/10.17605/OSF.IO/JE2DY). PubMed, Scopus, Web of Science, arXiv, and ClinicalTrials.gov were searched to March 2026. Eligible studies applied generative AI approaches to CNS-relevant targets and reported experimental wet laboratory validation. Titles and abstracts were screened independently by three reviewers; full-text screening, extraction, and quality appraisal were performed independently in duplicate. Findings were synthesised narratively; no meta-analysis was performed. Results Six studies met the inclusion criteria from 374 identified records. All performed in vitro validation; two additionally provided in vivo evidence. Despite CNS-focused objectives, one of six experimentally validated blood–brain barrier permeability. Three studies confirmed their primary computational predictions experimentally, whereas three showed partial confirmation. Validation rates, calculated as confirmed active compounds divided by compounds tested, ranged from 26 to 100 percent; these rates reflect differing selection stringency and are not directly comparable. Publicly accessible code was unavailable in four of six studies. Conclusions Among studies reaching experimental validation, blood–brain barrier confirmation, in vivo evidence, and code availability remain the exception. Because eligibility required reported validation, these findings describe successful translations and cannot estimate failure rates; they nonetheless indicate that validation depth, rather than molecular generation, is the current constraint in this field.
Source: ChemRxiv
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7213471277