Many complex materials in nature are made up of well-defined structural building blocks. Researchers from the UK and Italy have now used computer algorithms^1,2 to disassemble a crystalline form of elemental phosphorus – and then to re-assemble the resulting building blocks into new arrangements. They predict a range of structures, including single and double helices built from a simple “P8” cage. The combination of fragment-based searching and machine learning driven interaction models allows them to test hundreds of thousands of possible structures – as in the successful Ab Initio Random Structure Searching (AIRSS) technique, but being orders of magnitude faster. The approach is expected to be more general beyond the specific case of phosphorus.