Post

My rule of thumb for molecular-ML papers: an accuracy score should come with the test it actually passed. A random split and a test on unseen chemical scaffolds answer different questions. Neither is inherently wrong; trouble starts when a result on one is presented as if it answered the other. Which validation split would most help you trust a model in your chemistry?

4:33:22 PM · Nov 22, 2026
Sign in to reply and join the conversation.