AI tools in biomedicine are shifting from analysis to idea generation

In short: Biomedical researchers are increasingly using AI not just to analyze results, but to suggest what to test next and help design new drugs.

What's going on

AI has long helped scientists sort through large medical datasets. Now many research groups are training AI systems to generate hypotheses, which are educated guesses about what might be true, and then point to the best experiments to run.

One example is training AI on data from millions of human cells. The goal is to predict how genes and proteins may interact inside a specific cell, based on patterns seen in other cells. You can think of it like learning from many past traffic maps to guess which roads are likely to connect in a new city.

Researchers are also using AI to prioritize what to study next, such as which genes, genetic variants (small changes in DNA), or chemical compounds look most promising. In drug discovery, reviews describe AI being used for early steps like picking a drug target, designing new molecules, estimating toxicity (how harmful something might be), and finding new uses for existing drugs. Some systems can even propose brand new drug candidates aimed at specific proteins.

Large review papers describe similar moves across biomedicine, including biomarker discovery (biological signs that can help track disease), image analysis, and multi-omics work. Multi-omics means combining multiple layers of biology data, like genes, proteins, and metabolites (small molecules in the body), into one picture.

What to watch

The main benefit is speed. AI can sift through data that is too big or messy for people to comb through by hand, which can shorten the time from raw data to testable ideas. But researchers also stress a limit: AI suggestions still need real lab and clinical testing, so progress will depend on how well these tools hold up when experiments check their work.

Source: NYTimes