Review of 'Machine Learning and Artificial Intelligence in drug repurposing – challenges and perspectives'

INTRODUCTION - 3rd paragraph: "(...) aims to reduce the cost and time of development,
additionally seeing a much lower rate of clinical trial failure. Drug repurposing thus has the potential to dramatically decrease the time safe therapies may take to reach patients in need of a treatment[13]."
It should be mentioned that savings in time-to-market, cost, and clinical trial success rates are realized only in the early stages of (re)development, which are the fastest and cheapest already. The costs and timelines for the Phase III program (the most expensive part by far)  are unchanged.


Section by Ziaurrehman Tanoli:
Paragraph #4: "In cross-validation, the XGBoost prediction algorithm within RepurposeDrugs demonstrated a significant correlation of 0.75 for single drug-disease associations and 0.56 for drug combinations."
It would be useful to briefly describe what "cross-validation" means in this context. Also, has RepurposeDrugs been validated against real-life data, i.e., to which extent can it retrospectively predict actual drug repurposing successes?

No comments on the sections by Lucía Prieto Santamaría and Judith Bernett / Markus List.

The section by Adrian Freeman does not add additional perspectives to the previous sections in its present form, and should be expanded with more concrete content.