“Of the 450 “novel” results given in the paper, 135 of these results were not novel at all; they were already listed in the online database UniProt.”
A Nature Communications paper used a Transformer to predict functions for about 450 unknown E. coli enzymes. 135 of the “novel” predictions were already in the database it trained on. One predicted a mycothiol synthase in a bacterium that does not make mycothiol. The paper got 22,000 views, and the correction sits on bioRxiv getting a fraction of that.