No (cross) product liability

I am not sure I have ever seen a bioinformatics patent lawsuit in France before the case reported on today: this may well be a first.

It is therefore particularly interesting to see how our judges grappled with such an exciting and complex topic. And the short answer is: pretty well it would seem!

The Codexis group is a leader in the field of biocatalysts. In particular, Codexis Mayflower Holdings LLC owns European patent No. EP 1761879, filed in June 2005 and claiming a priority of June 2004. The patent was granted in August 2013 and was not opposed.

Claim 1 of the patent is the following (the wording which will be important for the discussion below has been emphasized):

A computer-implemented method for identifying amino acid residues for variation in a protein variant library in order to affect a desired activity, said method comprising:
(a) receiving data characterizing a training set of a protein variant library, wherein the data provides activity and sequence information for each protein variant in the training set;
(b) from the data, developing a sequence-activity model that predicts activity as a function of amino acid residue type and corresponding position in a protein sequence,
wherein the sequence-activity model includes one or more non-linear terms, each representing an interaction between two or more amino acid residues in the protein sequence,
and wherein the sequence-activity model can distinguish amino acid residues that have a significant impact on the desired activity from those that do not; and
(c) using the sequence-activity model to identify one or more amino acid residues at specific positions that are predicted to impact the desired activity for variation to impact the desired activity,
(d) wherein at least one of the non-linear terms is a cross-product term comprising a product of one variable representing the presence of one interacting residue and another variable representing the presence of another interacting residue, and
(e) wherein developing said sequence-activity model comprises selecting one or more cross-product terms from a group of potential cross-product terms, wherein the selected cross-product terms are those cross-product terms representing true structural interactions that have a significant impact on activity.

Unusually, the main defendant is an individual, and even more unusually, a university Professor, Mr. O.

Mr. O happens to teach biochemistry, molecular biology and bioinformatics in Nantes; he was blamed by Codexis for his personal website offering a tool for predicting protein sequences, called ProSAR. Furthermore, Mr. O is a vice-president of the company Peaccel, which provides services to biotech and pharma companies, and which uses the ProSAR software.

The issue, as you have probably guessed, is that Codexis deemed that the ProSAR software infringes its EP’879 patent.

In 2015, Codexis had an internet bailiff’s report, and then an infringement seizure at the professor’s home (!) carried out. The source code for ProSAR was seized and placed under seal. Codexis then formally initiated infringement proceedings against Mr. O and his company. The defendants filed a nullity counterclaim.

In 2016, an expert was appointed by the judge in charge of case management to analyze the source code and provide a comparison with the patent claims. The expertise was concluded in a matter of months.

In 2018, the Paris TGI issued its decision on the merits, rejecting both the nullity counterclaim and the infringement main claim.

Codexis appealed, Mr. O and Peaccel counter-appealed, which now leads us to the judgment rendered by the Paris Cour d’appel on January 15, 2021. This second judgment fully confirmed the first instance decision.

Starting with the validity discussion, one first item of contention was whether the priority claim was valid.

Indeed, Codexis’ inventors had originally filed a U.S. provisional application in 2002, followed by three successive continuation-in-part applications CIP1, CIP2 and CIP3 in 2003 and 2004. The EP’879 claims priority to the third one, CIP3.

Whenever patent attorneys see “continuation-in-part” and “European priority claim” in a same sentence, they smell blood in the water.

And indeed, the infringement defendants argued that CIP3 is not the first application for the claimed invention, as the first application is rather CIP2. As a result, they said, the priority is invalid, since only a “first application” can give rise to a priority right under article 87 EPC.

This argumentation must sound perfectly clear to EQE candidates, but it may not be an easy objection to deal with for a court of law. However, in this case, the court nicely set out the test to be applied:

It must therefore be first determined whether the subject-matter of the main claim 1 of the EP’879 patent and of dependent claims 2, 3, 7, 10 and 12 can be directly and unambiguously derived by the skilled person from the teaching of application CIP2, or from only the parts of application CIP3 not present in application CIP2. 

Then, the court noted that the patentee did not dispute that features (a) to (d) were directly and unambiguously derivable from CIP2. On the other hand, they did not believe that feature (e) could be derived from the passages cited by the defendants. As a result, CIP3 was acknowledged as the first application for the invention, and the priority was found valid. Too bad for the defendants, as the application CIP2 itself was published between the priority date and filing date of the patent.

Mr. O and Peaccel further raised objections of lack of novelty based on two articles published in 2001.

The first article, “W et al.“, apparently disclosed a method similar to that of claim 1, but applied to peptides, and not proteins as required by the claim.

Let’s take a step back to molecular biology 101: peptides and proteins (also referred to as polypeptides) both designate polymers comprising amino acid residues. But the term “peptide” generally refers to a short polymer, whereas a “protein” is a long polymer. Paragraph [0024] of the patent mentions a “typical” minimum number of amino acid residues of 30, 50 or 100 in a protein. In W et al., only three amino acids were present in the analyzed peptides.

The defendants insisted that the allegedly infringing software ProSAR works on both peptides and proteins, and that the internet bailiff’s report and expertise proceedings were actually conducted based on a test peptide of only ten amino acid residues, but this was considered irrelevant to the issue of novelty by the court.

From a peptide to a protein… to a living organism? Would the modelling still work then?

As to the defendants’ argumentation based on the second article , “L. et al.“, it was found to be so lacking that the court did not even bother to address it in detail:

The object of this publication differs from that of the EP’879 patent since it does not disclose a computer-implemented method to identify amino acid residues for variation in a protein variant library in order to affect a desired activity, but a method for rationally designing a medicament, […] as not contradicted by the very partial translation of the L et al. article filed [by the defendants]. [They] unsuccessfully attempted to combine various excerpts from incomplete sentences […] [so that] it is not necessary to reply in detail to this technical argumentation. 

Next, inventive step.

Starting from the first article W et al. as the closest prior art, Mr. O argued that the disclosed process was obvious to adapt from short peptides to actual proteins. The court was not convinced, as the proposed secondary reference related to a different process of protein engineering without features d) and e) of claim 1.

A second inventive step challenge also starting from a document directed to peptide engineering failed for similar reasons.

The patent was thus upheld. Let’s now turn to the issue of infringement.

Mr. O first argued that the analysis of the software relied upon by the plaintiff was based on a ten-amino acid test peptide – and thus not a protein. However, the court rejected the argument, since the software clearly works similarly with peptides and proteins. Squeeze-type arguments do not always work, and in this case the peptide/protein squeeze raised by the defendants left the judges unimpressed.

But another defense hit the mark, both in first instance and on appeal. Features d) and e) of claim 1 make reference to “a cross-product term comprising a product of one variable representing the presence of one interacting residue and another variable representing the presence of another interacting residue”In the ProSAR tool, a different calculation is performed, based on quadratic, term-to-term products.

Codexis’ technical expert acknowledged that there is a well-accepted mathematical definition of “cross-product” (“produit vectoriel” in French) but then submitted that a broader interpretation of this expression should be adopted in view of the description of the patent, which would cover any non-linear term expressing the interaction of two residues in the polypeptide. The court did not agree that a definition of cross-product different from the conventional one could be found in the patent.

The identification of this difference between ProSAR and claim 1 of the patent was not offset by other contextual arguments, such as allegedly incriminating statements made by Mr. O that he had been inspired by articles published by Codexis. The identified difference was sufficient to conclude that there was no literal infringement.

Codexis had another shot, based on the doctrine of equivalents. They claimed that the quadratic terms in the ProSAR software are equivalent to the cross product of EP’879.

Again, the court was unsympathetic to this view:

[…] Therefore, the equation of step 6 of the expert report and the one of paragraph [0101] of the description of the patent are neither identical, nor mathematically equivalent, and it can thus not be stated […] that the ProSAR tool is identical to the cross-product of the patent in terms of function and effect. 

Thus, it is not demonstrated by Codexis that the selection of quadratic terms implemented in the ProSAR tool fulfils the same function as the selection of terms of the cross-product in the patented method, namely the identification of interactions between acid amino residues which have an impact on activity, in order to achieve a same result, namely predicting the activity of new variants. The expert report does not confirm […] that the sole purpose of the genetic algorithm implemented in the ProSAR tool is to allow an efficient selection of non linear terms. 

It is slightly unfortunate that the judgment does not contain a more detailed discussion on this last aspect. Assessing infringement by equivalence in France requires determining what the function of the allegedly equivalent means is, but this determination is seldom straightforward. In the present case, based on the judgment itself, it is not easy to understand what the exact respective roles of the term-to-term product and the cross-product are in the process – and why.

Other than that, the quality of this ruling seems quite remarkable, given the high complexity of the technical field.

CASE REFERENCE: Cour d’appel de Paris, pôle 5 chambre 2, January 15, 2021, Codexis Mayflower Holdings LLC & Codexis Inc. v. Bernard O. & SASU Peaccel, RG No. 18/15295.

One thought on “No (cross) product liability”

  1. Since there seems to be an overlap between CIP 2 and CIP 3 it could have been an opportunity to introduce the notion of partial priority in French case law.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.