Patenting biomarker-related AI / machine-learning innovations

Shroyer and Michalek decisions

In Ex parte Shroyer, the invention pertained to methods for determining the likelihood of survival in patients with cervical squamous cell carcinoma by detecting concentrations of a particular keratin biomarker (KRT17). The specification described the rationale behind selecting this keratin as clinically significant, along with various immunohistochemical steps for measuring expression in fixed tissue samples.

The PTAB concluded that the claims, in essence, recited a correlation between naturally occurring keratin levels and survival outcomes. Although the specification described the potential advantage of evaluating KRT17 to inform clinical decisions, the PTAB was not persuaded the claims added any inventive concept. Importantly, the PTAB found that simply measuring and comparing KRT17 levels to a control constituted observation of a natural correlation, and the claims did not require a concrete application that could transform them into patent-eligible subject matter.

Similarly, in Ex parte Michalek, the claims focused on a system for classifying lung disease (non-small cell lung cancer and reactive airway disease) by measuring a large panel of biomarker concentrations and feeding the results into a machine-learning model, such as a support vector machine or an AdaBoost classifier. The specification described collecting patient data, normalizing the data, and using computational models trained on the normalized data to detect disease presence. The PTAB concluded that the claims recited either laws of nature (i.e., the correlation of biomarker patterns to disease states) or abstract ideas (as in the mathematical computations required for machine learning). Further, the PTAB found that incorporating steps of storing and analyzing data with general-purpose hardware did not sufficiently transform the judicial exception into a patent-eligible application. The specification’s explanation of the classification algorithm and how it was used did not provide a technological improvement beyond the mere application of a known mathematical model to a set of natural measurements.

In both Shroyer and Michalek, the specific passages in the specifications detailing how particular biomarkers were identified, or how a given machine-learning algorithm was configured, proved insufficient to surmount the “inventive concept” requirement. The PTAB repeated the principle that mere recitation of natural correlations, accompanied by routine or conventional detection steps, does not pass muster under Mayo and Alice. Where the claims did not recite transformative use or a practical application beyond the correlation itself, the examiner’s eligibility rejections were affirmed.

The Alice/Mayo test applied to Shroyer and Michalek

Under the Alice framework, as modified by the Supreme Court and subsequent United States Patent and Trademark Office guidance, examiners and the PTAB engage in a series of steps to evaluate subject matter eligibility:

• Step 1: Does the claim fall within a statutory category (e.g., process, machine, manufacture, composition)? The claims typically do, and Shroyer and Michalek were no exception. Per usual, the PTAB nodded to Step 1 and moved on to the more substantive judicial exceptions analysis of the next steps.
• Step 2A (Prong One): Does the claim recite a law of nature, natural phenomenon, or an abstract idea? In Shroyer, the PTAB found that correlating KRT17 levels with patient prognosis is essentially a law of nature. In Michalek, the underlying correlation of certain biomarkers with disease states also fell within natural laws, or, to the extent that classification steps dominated, the PTAB identified them as abstract ideas in the sense of mathematical algorithms. 
• Step 2A (Prong Two): Does the claim “integrate” the exception into a practical application? The PTAB concluded, in both Shroyer and Michalek, that the steps of measuring or analyzing did not transform the claims into something beyond the natural correlation or abstract computation. The claims merely instructed the practitioner to observe biomarkers or process data, without reciting a specific application that could render the overall invention patent eligible.
• Step 2B: Does the claim add “significantly more” than the judicial exception itself, i.e., an inventive concept? Again, the PTAB found that standard immunohistochemical assays, data-gathering, or commonplace machine-learning steps were not enough to supply an inventive concept.

Contrast with Ex parte Donovan

In Ex parte Donovan, the PTAB reversed a § 101 rejection of claims for diagnosing and predicting a response to breast cancer therapy by measuring multiple biomarkers (including phosphorylated HER2) and feeding the measurements into a machine-learning model. The PTAB found that the Donovan claims satisfied the Alice test at Step 2A, Prong Two, because they integrated the judicial exception into a practical application.  

Specifically, the claims required not only observing and measuring biomarkers, but also using specific hardware (such as an immunofluorescence imaging device and a computing device) and a machine-learning application configured to generate an effective treatment model using linear discriminant analysis. The claims further required generating, via a treatment module, an output data object that included information indicating a course of treatment for the subject, explicitly excluding treatments where the probability of resistance was above a given threshold.  

The PTAB favorably noted the claims included “additional elements sufficient to integrate the judicial exception into a practical application,” highlighting the orchestrated steps requiring specific interoperation of hardware and specially configured computing modules that generated effective treatment models and courses of treatment for cancer. The PTAB also observed that the specification described an integrative platform relying on machine learning to combine clinical data with quantitative biomarker characteristics, and that the claims were not merely abstract or routine, but instead recited a specific, practical application that transformed the nature of the claim into patent-eligible subject matter.  

Thus, unlike Shroyer and Michalek, where the claims effectively ended at “detect, compare, and conclude,” Donovan recited additional details and a concrete application—namely, generating a specific treatment recommendation—supported by a particularized hardware and software configuration. This was sufficient, in the PTAB’s view, to raise the Donovan claims beyond a mere judicial exception and render them patent eligible. 

Suggestions for improved drafting

In light of these decisions, applicants are well advised to draft claims that articulate how detected biomarkers or data correlations bring about a specific, technological application or improvement. When diagnosing diseases or evaluating biomarkers, it is not sufficient to merely observe and analyze a natural relationship. Instead, claims should recite further active steps that apply the discovered correlation in a meaningful way—e.g. triggering a particular type of medical intervention or adjusting a device in an inventive manner. 

Applicants should also consider reciting specialized technical configurations, for instance, by describing custom hardware or unique software architectures that do more than apply generic algorithms to data. Incorporating language indicating a transformation of the underlying sample or integration of real-time feedback can likewise help demonstrate a practical application.

Moreover, the specification itself must provide details beyond “why this biomarker matters.” It should underscore how the invention improves over conventional diagnostics, describing not merely the correlation but also how that correlation is harnessed to confer a new or enhanced capability. Showcasing improved sensitivity, throughput, or concrete medical actions can help evidence the invention is more than a bare mental process or natural phenomenon.

Suggestions for improved responses to examination reports

When responding to rejections under § 101, an applicant should carefully map claim elements to each prong of the Alice test. Demonstrate in detail how the claims do not merely recite a natural correlation or a disembodied algorithm. Identify any inventive steps that go beyond conventional laboratory or computational processes—perhaps highlighting advanced imaging, novel instrumentation, or a sophisticated software pipeline that yields new functionality. Also consider adding claim limitations that specify how the data will be used in a treatment protocol (for instance, choosing a particular therapeutic agent or dosage), thereby converting the correlation into a practical benefit.

Other strategies can also increase the likelihood of overcoming § 101 rejection. Submitting evidence, such as declarations or experimental data, is an effective tool for illustrating the real-world improvements or surprising technical achievements. Drawing on analogies to successful PTAB outcomes, like Donovan, can furnish persuasive arguments demonstrating the claims operate at more than the “observe-and-conclude” level. By tying each individual step to a coherent, inventive technical solution, applicants stand a better chance of distinguishing their claims from those that merely recite fundamental correlations or computations.  

Do not be discouraged by the challenges brought by § 101 or related subject-matter issues that may arise during foreign prosecution. Innovative technologies require innovative patent strategies. Our clients deserve protection for their life-changing biotech inventions.

Suggestions for counterpart foreign prosecution

When entering applications into foreign jurisdictions, additional care should be taken to avoid claim language that may be construed as methods of medical treatment. In numerous jurisdictions, including Canada, there are patent-eligibility standards relating to diagnostic or medical treatment methods, in addition to those related to laws of nature, natural phenomena, and abstract ideas. 

For example, in Canada claim language reciting “method of administering” or “obtaining a sample” is routinely rejected as defining non-eligible patent subject matter. Hence, applicants are well advised to seek guidance when filing corresponding applications in other jurisdictions. Specifically for Canada, the Canadian Manual of Patent Office Practice (MOPOP) currently has a set of draft proposals to amend the relevant guidance sections, specifically chapters 17 and 23.

To minimize risk of additional subject matter rejections during counterpart foreign prosecution, an applicant should take additional measures when drafting applications or during examination. 

When diagnosing diseases or evaluating biomarkers, include details on physical manipulation of biological samples and the technical configurations or reagents involved in sample treatment or analysis. Where diagnostic methods claims are presented, avoid reciting active medical intervention steps or replace with passive language. As well, “use claims” or “Swiss-style use claims” should be presented where appropriate. Fortunately, many US claim sets can be directly converted into jurisdiction-specific claim formats, with minimal additional statements placed on record at the patent offices.

For more information, please contact your IP professional at Norton Rose Fulbright Canada LLP.

For a complete list of our IP team, click here.