Structural genomics is biotechnology's newest thing, apparently—we said as much as far back as June 1998 (Nat. Biotechnol. 16, 595), and Nat. Struct. Biol. is devoting an entire supplement to it in November. Private investors have now caught on: they recently gave US startup Structural GenomiX $40 million. At least six startup companies are now basing themselves on “high throughput structuring”—producing proteins in a soluble form and preparing them for analysis by X-ray crystallography or NMR spectroscopy on an industrial scale. These companies claim they will be solving around 1000 structures a year. But building structural genomics businesses will not be easy.

Few drugs have sprung from protein structural studies, but that probably doesn't matter. It is precisely because structural genomics is not easy that there is any business opportunity at all. The difficulty is in working out exactly where that opportunity might be.

The database business model, for instance, worked well for sequence data but can it serve for structures? Probably not. Comprehensive databases would be elusive: the seven US NIH-funded pilot projects using microbial genome information have already indicated that the structures of proteins from only 5–20% of cloned genes may be readily solvable. In any case, it seems likely that pharmaceutical companies will want to characterize drug targets of their choosing rather than trawling databases of proteins of unknown relevance. Their targets could emerge from population genomics (which will define the frequency of relevant genetic alleles in the population) and by functional genomics (which will help identify potentially cross reactive structural motifs from non-target proteins). But structural information appears to be too expensive to be the first port of call for drug discovery and development.

It costs $200,000 to solve a protein structure with current approaches. While the start-ups say this may fall to $20,000 per protein with high-throughput methods, the most expensive structures of all to solve are likely to be those of the proteins of greatest commercial interest. A high proportion of pharmaceutically interesting targets are likely to be proteins embedded in membranes. Membrane proteins are difficult to synthesize and require heroic efforts to crystallize and model: the Protein Databank holds ∼1,000 unique protein structures; only 26 are membrane proteins.

But perhaps the most compelling risk for structural genomics business models is the threat of commoditization. Companies in the field must compete not only with each other but also with large companies and with academia. The UK's Wellcome Trust and a consortium of around 16 pharmaceutical companies plans to fund a substantial increase in the flow of novel structures into public databases. The NIH's $125 million Protein Structure Initiative, starting in earnest later this month, also aims to fill the protein structure space with 10,000 representative motifs. Startups surely will have to be exceptionally smart, agile and fast.