Recent advances in protein structure prediction using machine learning approaches such as AlphaFold and RosettaFold presage a revolution in structural biology. Genome-wide predictions of protein structures are providing unprecedented insights into their architecture and intradomain interactions and applications have already progressed towards assessing protein complex formation across whole proteomes. Here we present detailed analysis of the sorting nexin proteins that contain regulator of G-protein signalling domain (SNX-RGS proteins), providing a key example of the ability of AlphaFold to reveal novel structures with previously unsuspected biological functions. These large proteins are conserved in all eukaryotes and are known to associate with lipid droplets (LDs) and sites of LD-membrane contacts with key roles in regulating lipid metabolism. Previous studies indicated they possess five domains, including an N-terminal transmembrane domain that anchors them to the endoplasmic reticulum, an RGS domain, a phosphollipid interacting phox homology (PX) domain and two additional domains named the PXA and PXC domains of unknown structure and function. Here we find that the distant PXA and PXC domains in fact fold into a single unique structure that notably features a large and conserved hydrophobic pocket. The shape of this pocket strongly suggests a role in lipid or fatty acid binding and we propose that these molecules represent a new class of lipid transfer proteins.