Most predicted drug pockets in the apicomplexan dark proteome are artefacts of non-globular structure.

Abstract

Structural genomics is used to find drug targets among the proteins a genome leaves unnamed: a protein with no sequence homologue but a confidently predicted fold looks both novel and tractable. I show that in Plasmodium falciparum this reasoning fails four times before it reaches a target list. Of 1017 confidently modelled hypothetical proteins, 316 appear structurally novel; searching experimentally determined structures reassigns 105, so a third of the novelty is a gap in the reference database. Against those 105 as a matched control, the remainder differs in chromosomal distribution, secondary structure and essentiality by no measurable amount. Strict annotation from two sources leaves 99 genes of 210. Of these, 59 carry a pocket above the usual druggability threshold, but most models are extended rather than globular and a rod has no interior; requiring a compact fold and a pocket walled by several sequence segments leaves 10, two of them essential with a confident lining. Model confidence detects none of this. Across 15 apicomplexan genomes the extended fraction varies six-fold and tracks low-complexity content, a long-recognised feature of these proteomes whose consequence for pocket prediction has not been quantified. Screen reliability is an organism-level property, measurable in advance.