A roadmap that works every time is still trial and error and does not enable a functional genus claim

2023-09-20Fed. Cir.

Baxalta Incorporated v. Genentech, Inc., 81 F.4th 1362 (Fed. Cir. Sept. 20, 2023) (precedential) (Moore, C.J., joined by Clevenger and Chen, JJ.), No. 2022-1461, affirming Baxalta Inc. v. Genentech, Inc., 579 F. Supp. 3d 595 (D. Del. 2022) (Dyk, J., sitting by designation)

United States Patent No. 7,033,590 claims in claim 1 "an isolated antibody or antibody fragment thereof that binds Factor IX or Factor IXa and increases the procoagulant activity of Factor IXa," a functional genus for treating hemophilia A. The inventors made antibodies by the prior art hybridoma technique, screened thousands, found that 1.6 percent increased procoagulant activity, and disclosed the sequences of eleven. Baxalta sued Genentech over Hemlibra, a humanized bispecific antibody, and the district court, Circuit Judge Dyk sitting by designation, granted summary judgment that claims 1 through 4, 19 and 20 are invalid for lack of enablement under Section 112(a) of Title 35 of the United States Code. The Federal Circuit affirmed in a precedential opinion by Chief Judge Moore, finding the facts "materially indistinguishable" from Amgen Inc. v. Sanofi: millions of candidate antibodies, eleven disclosed, and a written description that directs the skilled artisan to immunize mice, form hybridomas, and test for binding and activity, the same iterative, trial-and-error process the inventors followed, with no common structural feature or other quality that predicts which antibodies work. Baxalta's argument that the process yields a claimed antibody every time it is run did not help; the same was apparently true in Amgen. The court saw no meaningful difference between the Wands "undue experimentation" standard and Amgen's "[un]reasonable experimentation" and held that Amgen did not disturb Wands.

What it changesMachine learning specifications routinely teach the same roadmap: assemble candidate architectures or hyperparameter settings, train each, evaluate against a metric, and keep the one that meets it. A claim to every model that achieves the metric, supported by a few worked examples and no disclosed property that predicts which configurations succeed, is the antibody claim in this case translated into software, and the argument that the training loop reliably produces a qualifying model each time it is run was made here and lost. Draft the independent claim to the mechanism that makes the model work, such as the architecture, the loss term, the data transformation or the training schedule, and put in the specification the reason those features produce the result, so the claim rests on a disclosed quality common to the embodiments rather than on a screen. For issued claims defined by a performance threshold, review the specification for that common quality before asserting them.

More on Written description, enablement and claim scope

← All recent developments · Search all developments

Burbage Law is the practice of Jon Burbage, a registered patent attorney. 342 United States patents have issued from applications he wrote; the full list is on the Patents page. More about Jon Burbage.

This is general information about a published decision, agency guidance or piece of legislation. It is not legal advice and does not create an attorney-client relationship. Items concerning offices outside the United States are reported for information only and are not counsel on the law of those countries.