This matters because protein structure prediction is increasingly being run on sequences from non-model organisms where we have little prior knowledge. Erroneous sequences from genome misassembly can silently percolate through prediction models into structure databases. [8/9]