Patent directed to determining DNA methylation using machine learning survives Alice motion-to-dismiss rabbit hole

Much commentary has been devoted to the patenting of claims reciting the use of machine learning and artificial intelligence. These patents are now being litigated and courts are grappling with the application of patent law jurisprudence to this new technology. In Take2 Technologies Ltd. v. Pacific Biosciences of Cal. Inc., the district court for the Northern District of California recently analyzed whether the plaintiffs’ asserted patent, US Patent No. 11,091,794 (the '794 patent), was invalid under 35 USC § 101.

The ’794 patent is directed to improvements in genetic sequencing of methylated nucleotides. DNA methylation, which occurs when methyl groups are added to cytosine bases of eukaryotic DNA, influences gene expression and genome stability. Abnormal DNA methylation patterns are frequently associated with cancer.

Ordinarily, in order to assess methylation, the DNA sample must be chemically pretreated with bisulfite and then undergo PCR amplification prior to sequencing. The '794 patent, however, purports to disclose a method that sidesteps the need for chemical pretreatment and PCR amplification.

At a high level, the ’794 patent describes a four-step method: (1) collecting optical pulse information for each nucleotide in a DNA fragment; (2) creating an input data structure for the optical pulse information associated with each nucleotide in a specific portion of the DNA fragment; (3) feeding this input data structure into a machine learning model; and (4) using the machine learning model to determine the presence of a DNA modification, such as methylation.

The plaintiffs, The Chinese University of Hong Kong and Take2 Technologies, sued Pacific Biosystems (PacBio) for patent infringement, asserting that PacBio infringed at least independent claim 1. PacBio filed a motion to dismiss pursuant to Fed. R. Civ. P. 12(b)(6), arguing that the ’794 patent claimed ineligible subject matter and was invalid under § 101. More specifically, PacBio argued that the ’794 claimed nothing more than “a statistical model for calculating an allegedly better prediction as to whether a nucleic acid base is modified.” PacBio relied, in part, on In re Board of Trustees of Leland Stanford Junior, 991 F.3d 1245 (Fed. Cir. 2021), which held that a claim "directed to the use of mathematical calculations and statistical modeling" is abstract and runs afoul of § 101.

The district court applied the Supreme Court’s two-step patent eligibility framework, the Alice test, to evaluate the ’794 patent. At Step One, the court must determine whether the claims as a whole, considered in light of the specification, are directed to ineligible subject-matter. Here, the court agreed with the plaintiffs that the claims were directed to an abstract concept under Alice Step One. Following Federal Circuit precedent, the court found that the claimed steps of gathering information, analyzing it via mathematical techniques, and reporting or displaying the results is abstract.

More specifically, the court first noted that the specification indicated advances in the “deduction that using specific metrics available from a DNA sequencing read,” which included interpulse duration, “could lead to a more accurate prediction of base modification.” Thus, the claims were directed to “retrieving specific data inputs from a sequencing read[,] . . . creating a data structure based on the data, and inputting the data structure into a model trained to predict the presence of a nucleotide base modification based on the relevant inputs, [] thereby determining the existence of the modification.” Thus, according to the court, the claims were not directed to technological improvements in detecting base modifications but rather to “an algorithmic improvement within the field of research on using optical pulse data to detect base modifications.” While the claim uses the claim term “model,” the court’s analysis interpreted “model” as “neural network model.”

At Step Two, the court must "consider the elements of each claim both individually and as an ordered combination to determine whether the additional elements transform the nature of the claim into a patent-eligible application." The Federal Circuit has provided further guidance, instructing that the inquiry is whether the claims involve “something more” than well-understood, routine, or conventional activities previously known to the industry. This is also known as the “inventive concept.” Here, the court concluded that there were indications of an inventive step, or “something more” beyond the abstract idea. The court rejected the defendant’s argument that adding new parameters to a statistical model is insufficient to survive Alice Step Two. It pointed to other claim elements that the defendant’s argument did not address, such as a “window of the nucleotides sequenced" and the placement of a “nucleotide at the target position in the respective window.”

Moreover, whether the elements of a claim or the claimed combination are well understood, routine, and conventional is a factual question. Accordingly, if there is a genuine dispute of fact, then a motion to dismiss on the basis of ineligibility is foreclosed. The complaint also made several specific allegations regarding industry praise and characterizing the invention as an “unconventional concept.” These allegations must be taken as true by the court in evaluating a motion to dismiss. Consequently, the court denied PacBio’s motion to dismiss.

The court’s ruling offers a few takeaways for plaintiffs and defendants litigating patents that claim elements of machine learning or neural networks, especially in the DNA sequencing space. Plaintiffs should consider drafting complaints to include as many factual assertions as possible related to industry praise and the unconventionality of the invention to better shield the patent from any invalidity challenge at the motion to dismiss stage. In turn, defendants should consider addressing all of the elements of any challenged claim and highlighting the lack of inventive concept of all limitations.

Section 101 challenges remain a significant barrier to such claims, whether at the motion to dismiss stage or later in the case. It also remains to be seen how courts will address claim limitations directed to the training of the machine learning model – and what is sufficient to demonstrate an inventive concept. We note that although the plaintiffs’ patent has survived Alice at a motion-to-dismiss stage, molecular diagnostic patents have been repeatedly invalidated at the Federal Circuit. Thus, careful litigation planning and strategy is essential.

Find out more about the implications of this ruling by contacting either of the authors or your usual DLA Piper relationship attorney.