The informal narrative generally used to describe the requirements for obtaining a patent is that an inventor must submit a patent application that discloses an invention which is novel, non-obvious, and useful. In practice, there is an additional requirement: the invention must be “patent-eligible”.
Section 101 of Title 35 of the US Code defines subject matter eligibility for patentability as follows: “Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor…”. In practice, the US Supreme Court has long interpreted patent eligibility as excluding abstract ideas, natural phenomena, and laws of nature. While it was not the first case to do so, as one example: the US Supreme Court’s opinion on Diamond v. Chakrabarty clearly set out these boundaries. In that case, a genetic engineer (Chakrabarty) working for General Electric had developed a bacteria capable of breaking down crude oil, which he proposed as useful for cleaning up oil spills. The Supreme Court ruled in Chakrabarty’s favor, arguing “While laws of nature, physical phenomena, and abstract ideas are not patentable, respondent’s claim is not to a hitherto unknown natural phenomenon, but to a nonnaturally occurring manufacture or composition of matter…”
US SUPREME COURT RULINGS
A delicate balancing act
Recently, a set of four US Supreme Court rulings has clarified and - arguably - expanded the boundaries of what will be interpreted as non-patent eligible going forward:
1) In Bilski v. Kappos the Court invalidated patent claims on an investment strategy, announcing it supported a “high enough bar” on patenting abstract ideas that it would not “put a chill on creative endeavor and dynamic change.” The claimed invention in this case was a system for buyers and sellers in energy markets to hedge against the risk of price changes; the patent application included a claim over a mathematical formula that can be applied to minimize risks from market fluctuations. The Court ruled that this claim was an abstract idea and therefore patent-ineligible.
2) In Mayo v. Prometheus, the Court invalidated patent claims on methods of using genetic variation to guide pharmaceutical dosing, expressing concern that “patent law not inhibit further discovery by improperly tying up the future use of laws of nature.” The claimed invention in this case was a process by which physicians could measure patients’ metabolite levels in order to determine the risk of thiopurine drug administration. The Court held that the relationship between metabolite concentrations and thiopurine risk is a law of nature and therefore patent-ineligible. Moreover, the Court argued that the steps outlined in the patent claims for administering and reading the test did not add anything that transformed the process into patentable subject matter.
3) In AMP v. Myriad the Court ruled to invalidate a subset of Myriad’s gene patent claims, arguing that such patents “would `tie up’…[genes] and…inhibit future innovation premised upon them.” The claimed invention on this case was human genes correlated with risks of breast and ovarian cancer. One technical detail that is critical to understanding the AMP v. Myriad case is that two types of nucleotide sequences were at issue: naturally occurring genomic DNA (gDNA), and complementary DNA (cDNA), the latter of which is produced in a laboratory using gDNA as a template. The Court decision drew a distinction between these two types of sequences: “A naturally occurring DNA segment is a product of nature and not patent eligible…but cDNA is patent eligible because it is not naturally occurring.”
4) In Alice Corp v. CLS Bank the Court invalidated patent claims based on similar arguments. The claimed invention in this case was a scheme to mitigate “settlement risk,” or the risk that one side to a settlement agreement will not meet their obligations.
The patent holder claimed a method for exchanging financial obligations, as well as a generic computer system and code to carry out the obligations. The Court held that this method of exchanging financial obligations was an abstract idea and therefore not patent eligible. They further found that the additional claims, which tie the method to the use of generic computer systems, did nothing to transform the idea into patent eligible subject matter.
What is eligible for patent?
Taken at face value, the implications of these four rulings are incredibly broad. The Court has taken four technological areas - business methods, medical diagnostics, DNA, and software - and deemed that broad sets of inventions in those areas should no longer be eligible for patent protection. With the caveat that my background is in economics, not in law, my interpretation of these rulings is that the Court is relying on the Section 101 patent eligibility definition as a way of “carving out” certain technologies where they feel that the social costs of patents outweigh the social benefits.
For example, the Court’s Mayo v. Prometheus decision argued that the patenting of abstract ideas may tend to impede innovation more than it encourages it. This is of course a theoretical possibility, and if such decisions were citing or otherwise building on rigorous empirical evidence that the social benefits of patents in those areas were indeed outweighed by the social costs, I could see the logical case for such decisions. However, my read of the available empirical evidence is that there is essentially no strong empirical foundation either for or against that view.
Legal scholars have analyzed how the Court’s reasoning over patent eligibility under Section 101 has progressed over the course of these cases. For example, in Bilski v. Kappos the Federal Circuit ruled that subject matter was not patent eligible unless it passed the “machine-or-transformation” test, which requires that a process is not patentable unless it is tied to a particular machine or transforms an article to another state or thing. Lemley et al. (2011) as well as other scholars have argued that this test is flawed, and consistent with that view the Court later argued that while the machine-or-transformation test is a clue to patentability, it is not dispositive to the question. As a second example, in Alice Corp v. CLS Bank the Court used a two-step test derived from its reasoning in the Mayo v. Prometheus decision: the Court first determined whether the claims were directed at one of the excluded classes; if they are, then the Court next determined whether the claims contain an “inventive concept” that “produces something more than an attempt to claim the prohibited subject matter” (see Burk 2014).
While these legal analyses have been extremely valuable in examining the reasoning used by the Court, a valuable complement to such analysis is the development of empirical evidence on what I view as the key question underlying these rulings: do patents tend to impede innovation more than they encourage it for these technologies?
Below, I summarize some ongoing work that is starting to develop evidence on this question in one context - the AMP v. Myriad case of gene patents - to give a sense of the direction that I hope future empirical research will develop to investigate the economic questions in this area.
THE AMP V. MYRIAD CASE
Do gene patents impede follow-on innovation?
The private firm Myriad Genetics was granted patent rights on human genes correlated with risks of breast and ovarian cancer. In 2009, the American Civil Liberties Union (ACLU) and the Public Patent Foundation filed suit against Myriad, arguing that many of Myriad’s patent claims were invalid on the basis that DNA should not be patentable. In June 2013 the US Supreme Court unanimously ruled to invalidate a subset of Myriad’s gene patent claims, arguing that such patents “would `tie up’…[genes] and… inhibit future innovation.” In terms of relevance to Section 101 in particular, the AMP v. Myriad decision argued that “[g]roundbreaking, innovative, or even brilliant” discoveries of natural phenomena should be patent-ineligible, because patents “would `tie up’ the use of such tools and thereby inhibit future innovation premised upon them.” As discussed by Rai and Cook-Deegan (2013), the Court decision essentially aimed to draw a line between patent-eligible and patent-ineligible discoveries based on the “delicate balance” between patents prospectively creating incentives for innovation and patent claims blocking follow-on innovation. In the end, as discussed above, the Court drew this line by ruling naturally occurring gDNA patent-ineligible, and nonnaturally occurring cDNA patent-eligible. Numerous legal scholars have argued that the distinction between DNA and cDNA is “puzzling and contradictory” (Burk, 2013) given that “both isolated sequences and cDNA…have identical informational content for purposes of protein coding” (Golden et al., 2013); in interviews, patent attorneys have expressed similar confusion.
While - consistent with the Court’s view - there has been widespread concern that patents on human genes may hinder follow-on innovation, as there was essentially no empirical evidence available to either support or refute that assertion. In a recent working paper, economist Bhaven Sampat and I set out to try to develop empirical evidence on whether patents on human genes have hindered follow-on innovation in practice.
WEIGHING THE EVIDENCE
How to protect innovation in genetics
Investigating how patents on human genes - or more generally, on other existing technologies - affect follow-on innovation requires addressing two key challenges. First, in most markets it is extremely difficult to measure follow-on innovation empirically. We have a sense that in many or most markets, innovation is cumulative in the sense that any given technology is often an input into subsequent technological change, enabling follow-on discoveries.
But measuring this enablement is very difficult in practice. Second, ideally we would have something akin to a randomized experiment, where some human genes were randomly patented and others were not, and then we could confidently attribute any difference in follow-on innovation across patented and nonpatented genes to a causal effect of the gene patents on follow-on innovation. In practice, inventors choose which genes to file patent applications for, and the patent office chooses which patent applications to grant patents to. Both types of selection raise the concern that any measured differences in follow-on innovation across patented and non-patented genes could reflect the selection of which genes were patented, rather than the causal effect of patents on follow-on innovation.
Our study aims to circumvent both of these challenges. To address the first - measurement - challenge, we take advantage of the fact that US patent applications claiming intellectual property rights over human genes are required to disclose the exact DNA sequences being claimed in the text of the patent.
By applying standard bioinformatics methods, these DNA sequences can be linked to gene identifiers, and these gene identifiers can in turn be linked to a variety of medical and scientific databases measuring follow-on scientific research and product development related to the human genome.
For example, gene identifiers are linkable to scientific publications in the PubMed (https://www.ncbi.nlm.nih.gov/pubmed/) database cataloging publications in the biomedical literature, to some datasets cataloging clinical trials in progress by both public and private drug development research organizations, and to datasets cataloging the availability of gene-based medical diagnostic tests. From a measurement perspective, this linkage from patent applications to the “bench” (scientific research) and to the “patient” (in terms of commercialized or under-development medical technologies) is remarkably complete.
Because we observe these measures of follow-on scientific research and product development for all human genes, this data construction is sufficient to tabulate a preliminary answer to our question: do patented human genes have more or less follow-on innovation? It turns out patented genes have higher levels of follow-on innovation. However, indicative of the selection bias challenge described above, it turns out that genes that will be patented in the future have higher levels of follow-on innovation even in the years prior to when these genes are patented.
This suggests that selection bias is a major concern, and that in particular the direction of selection is that patented genes look like they had higher potential for follow-on innovation even in the absence of their patents. This highlights the need for a research methodology that addresses this type of selection.
To address this second - selection - challenge, we develop two new quasi-experimental methods for estimating the causal effect of gene patents on follow-on innovation. First, we present a simple comparison of follow-on innovation across genes claimed in accepted and rejected patent applications. This method is valid if, conditional on being included in a patent application, whether or not a gene is granted a patent is as good as random. Consistent with this assumption, we document that patented genes look similar - in years prior to the patents being granted - to genes that are included in patent applications but not granted patents. Second, we develop a novel approach for estimating a causal effect of patents on follow-on innovation that takes advantage of the “leniency” of the assigned patent examiner.
While patent examiners have a uniform mandate, prior research has documented that in practice this mandate appears to leave patent examiners with a fair amount of discretion. We leverage this across-examiner leniency variation together with the fact that patent applications are quasi-randomly assigned to examiners, conditional on some covariates such as the application year and technology type, to approximate the following thought experiment: two equally high-quality gene patent applications come into the US Patent and Trademark Office at the same time, but one is more likely to be granted a patent because it is assigned to a more lenient examiner.
In contrast with the basic tabulations described above, both of our quasi-experimental approaches suggest that gene patents have not had quantitatively important effects on either follow-on scientific research or on follow-on commercial investments. These conclusions speak against the Supreme Court’s argument in AMP v. Myriad, and more generally suggest that - as best we can measure - gene patents may not have had negative effects on follow-on innovation.
A guide for future empirical research
Of course, this work is still preliminary. I highlight our study largely as an example of the type of empirical research that I would like to see more of in the future. Software patents may be good or bad for innovation, but rather than having policy or judicial decisions about software patents be based on theories or ideologies, these decisions would instead ideally be based on rigorous empirical evidence.
Our study provides an example of how standard datasets on patent applications and granted patents - which are publicly available from the US Patent and Trademark Office - can be combined with other data measuring innovation to develop quasiexperimental evidence on such questions.
The preceding is republished on TAP with permission by its author, Professor Heidi Williams, and by the Toulouse Network for Information Technology (TNIT). “Which discoveries should we patent?” was originally published in TNIT’s April 2017 newsletter.