Understanding the claims in this machine learning patent application

Question

I was looking at an American machine learning patent application the other day, and it detailed many specific parts. It goes through how they label the data to solve their problem, and how they train the model. Here they specifically mention that they used the method SVM to train their model.

However, a later section, starting on page 24, of the patent is called "claims". Here they claim "A method for training a machine learning algorithm or statistical inference model" to solve the problem they tried to solve using the way of labelling data they used. This sounds like they now trying to patent that way of labelling the data for said problem regardless of which ML model is used and not just SVM like they used. However, this way of labelling data predates their patent.

Since I may be reading the patent application wrong, I wonder how specific that patent would be if it is granted. Would that entail that nobody else can split their datasets into positive and negative datasets for identifying peptides in a commercial setting? Or, does it only apply if that team also uses SVMs?

Also, how should I read the claims section of the patent? Is it one claim with several parts (meaning it's one very specific claim)? Or, are they trying to be granted several separate claims? I see later claims are referring to earlier claims, but I am still having a difficult time completely understanding the those.

Answer 1

The linked document is a WO application. It is not yet a patent and may or may not get granted. Even if it is eventually granted in one or more countries, it is likely the claims will be changed and likely narrowed. The claims could end up being different in the different countries the applicant is applying to. I know a bit about machine learning and less about molecular biology. That said I'll try to provide some insight to the first claim which looks like the only independent claim.

1. A method for training a machine learning algorithm or statistical inference model to identify peptides that contain features positively associated with natural endogenous or exogenous cellular processing, transportation and major histocompatibility complex (MHC) presentation, that negates the influence of HLA MHC-binding and can be applied to any peptide regardless of its MHC restriction, comprising:

   (a) building one or more training data sets comprising a positive and a negative data set; wherein the positive data set comprises entries of peptide sequences identified or inferred from surface bound or secreted HLA/MHC/peptide complexes encoded by one or a plurality of different HLA/MHC alleles, and wherein the negative data set comprises entries of peptide sequences which are not identified or inferred from surface bound or secreted HLA/MHC/peptide complexes; wherein the training data further comprises a multiplicity of pairings between entries of the positive and negative data sets; and wherein each pair of said multiplicity of pairings comprises peptide sequences which: (i) are of equal or similar length,

   and

   (ii) are derived from the same source protein or fragment thereof,

   and/or

   (iii) have similar binding affinities, with respect to the HLA/MHC molecule which the positive counterpart is restricted, and (b) applying a machine learning algorithm or statistical inference model on said training data.

This is a fairly long claim with quite a few elements. Many people see long claims and assume they provide broad coverage, but in actuality, long claims are generally narrower than short ones. In order to infringe on a claim, you need to implement each and every step in the claim. If a claim has steps A, B, C and D and you only implement steps A, B and D you don't infringe the claim. The other claims are dependent claims. Dependent claims narrow the scope of the independent claim. If you avoid infringement of claim 1 in this application then you shouldn't have to worry about any of the dependent claims.

Answer 2

To supplement the excellent answer by @EricS - The claims do not limit themselves to SVM so it is not required in order to infringe these claims. This is fairly standard since there are many methods to get the result but some specific method should be presented in the specification.

A patent should explain at least one specific way of performing the invention in some detail. In the U.S. the requirement is to teach someone skilled in the art how to "make and use the invention".

If SVM was a particularity good choice in some invention it might be in a dependent claim as a fall back.

Eric mentions differences between the claim wording in the PCT and U.S. versions. Note the wording "according to any preceding claim" in the international version. Since there is more than one preceding claim this is called a multiply dependent claim - it depends from more than one other claim. This form is encouraged in the EPO and most of the world. In contrast, it is allowed but heavily discouraged (and financially burdened) in the U.S. so you do not see this structure in the U.S. version.