So How Is AI Drug Discovery Doing, Really?
Summary
Derek Lowe argues that AI drug discovery lacks evidence of clinical impact because current methods focus on modeling readily available data rather than generating the specific data required to improve Phase II success rates.
Clinical impact remains unproven.
Current AI applications show an absence of evidence regarding their ability to translate into meaningful clinical outcomes or improved patient results.
evidence of their clinically relevant impact is, so far, disappointingly limited
Phase II is the true metric.
Improving Phase II success rates is the only way to move the needle, as early-stage discovery costs are negligible compared to clinical trial failures.
Data availability masks utility.
Practitioners often model existing data because it is easy, but this fails to address the substantial data generation needed for real progress.
focus of AI in drug discovery must shift from doing what can be done
Biological complexity breaks machine learning.
The massive number of variables in assay conditions and biological systems makes cleaning and categorizing data for ML potentially impossible.
machine-learning your way to happiness with those numbers is very likely impossible
Benchmarks create illusions of progress.
Confounding factors and epistemic opacity in drug data mean that believing labels can lead to flawed benchmarks that do not translate to real-world use.
the limitations of labelling data and using labelled data are not understood
Highlights
Although a wide variety of AI methods have been developed, applied and benchmarked, evidence of their clinically relevant impact is, so far, disappointingly limited
the nature of the data at hand in drug discovery — its confounding factors, conditionality and epistemic opacity — may not be understood by everyone
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.