Open source has contributed wonderfully to collaboration in bioscience. But I’ve got problems with the argument, made by the Open Bioinformatics Foundation and others, that all federally funded bioinformatics software should be released as open source. Here’s why.
Requiring open source exchanges one form of inflexibility for another. Patent commercialization at technology transfer offices has failed bioinformatics software. The answer, though, is not to impose another inflexible system in its place; we benefit from a diversity of source management approaches. If it comes to open source purity versus dissemination freedom, I prefer freedom. Some of us involved in technology transfer — including Patrick Jones at the University of Arizona and Charles Williams at the University of Washington — have worked to encourage source-available and proprietary models. Our “second generation” methods are well suited to bioinformatics software and data deployment, are responsive to the practice of publicly funded science, and give developers distribution choices consistent with their understanding of community needs.
Current proposals specify the narrow open source protocols of the Free Software Foundation and the Open Source Initiative. Valuable as FSF and OSI protocols are, there are meaningful source-available alternatives. For instance, Phrap source is distributed by the University of Washington without charge to non-profits. For-profits pay modest fees that help support the lab. The UW source distribution doesn’t meet narrow definitions of “open source” because, among other things, it favors academics and regulates redistribution. Yet the Phrap distribution is entirely consistent with public science — Phrap algorithms are published, academic researchers get ready access, and companies can deal with the lab or with a commercial developer such as Geospiza or Codoncode.
Similarly, the Genome Browser at the University of California, Santa Cruz, is freely available for use on the web, and we anticipate providing source with licensing protocols that support the NIH guidelines and creatively address differing interests of academic researchers, established biotech companies, startups and smaller companies, software distributors, and government agencies.
Proprietary models do some things well. Science benefits from competing theories, methods, and tools. Celera motivated a huge increase in the publicly funded effort to sequence the human genome. Proprietary development of software has shown a consistent ability to create innovative products. In contrast, popular open source protocols evidence a tendency to monoculture, focused on emulating proprietary codes and making local variations in existing open source platforms. Call it “fear of the fork.” We can fall into monopoly just as easily with enforced open source as any other way. It may be that proprietary codes and open source form a coevolutionary assembly. If so, then for open source to be effective, it actually needs complementary proprietary efforts.
Keep the separation of source and state. As an enterprise matures, it may reach a critical phase transition where it will be tempted to dominate an entire market. Government-mandated open source imposes a regulatory cathedral over the programming bazaar, and it’s disconcerting to see such a proposal arise with open source advocates. The “free” in free choice is as important as the “free” in free speech (and as desirable as the “free” in free beer). If bioinformatics open source dominates by law, then there will be one code to rule us all, and collectively we will lose much that we won’t easily win back — for scientific communication, for the interplay between public and industry science, and for development of progressive source management strategies.