Inside every cell, work is constantly underway: proteins communicate, genes turn on and off, and glitches in this system lead to disease. To find the culprit behind an ailment, scientists used to manually comb through research papers and databases, like a detective interviewing witnesses one by one. But there are so many connections that no one can handle it alone.

The new model, GLaDiGAtor, works like a smart graph: it builds a map where nodes are genes and diseases, and edges are known interactions. Then it learns from this map, as if from examples, and predicts where missing edges should be. The map also incorporates textual hints: disease descriptions and protein sequences, which the model reads as annotations on the nodes. This allows it to see not only direct connections but also hidden patterns.

This approach speeds up the search for drug-target genes. Instead of months of manual work, just a few hours of computation. And the more data goes into the map, the more accurate the predictions become. Perhaps soon we'll be able to foresee which gene will fail in a given disease and prepare treatment before symptoms appear.