Patenting AI Tools of Drug Discovery
Downstream Claims
Having drafted many patents on the AI aspects of AI tools of drug discovery, we now endeavor to draft “downstream claims” that capture the value of the downstream product that emerges from the commercial implementation of such AI tools in the drug development pipeline (e.g., a therapeutic agent that passes through preclinical and clinical study to emerge as a multi-billion blockbuster drug).
Product-By-Process Claims
As a gateway to the downstream claims, we draft “product-by-process” claims that impart the backend ingenuity of such AI tools to their frontend output. We use a patentable and structural characterization of the frontend output as the underpinning of a broad downstream claim that covers later stages of the drug development pipeline like target discovery and lead discovery.
PrimateAI Case Study
We use the patented disclosure of a leading variant classification tool called PrimateAI as a case study to demonstrate how our downstream claims, informed by by our product-by-process claims, advance PrimateAI to later stages of the drug development pipeline.
Target
Discovery
Lead
Discovery
Preclinical
Research
Clinical
Research
Process
Development
Manufacturing
& Testing
Drug Development Pipeline
PrimateAI’s patents
place it here.
Can we draft patent claims that place PrimateAI here?
How about here?
“Jump first and then think as much as you want!”
PrimateAI Case Study
PrimateAI Publication
We studied forward citations to PrimateAI’s publication in Nature Genetics to understand where and how PrimateAI is being discussed.
PrimateAI Scores
We found prevalent use of PrimateAI’s output called “PrimateAI Scores” for target discovery like identifying cancer driver mutations, Alzheimer's disease risk genes, and other loss-of-function variants.
Illumina’s AI Pact with AstraZeneca
Illumina’s AI Pact with AstraZeneca validates our findings that PrimateAI Scores are the “commercial speartip” of PrimateAI that are being used for drug discovery.
“Few things are harder to put up with than the annoyance of a good example.”
Patentable AI Output: Draft product-by-process claims that broadly transfers the patentable weight of AI’s backend intricacies to its frontend output.
Integrate Patentable AI Output into Target Discovery: Draft downstream claims that identifies a target that is structurally implied by the patentable AI product of 1.
Integrate Patentable AI Output into Lead Discovery: Draft downstream claims that produces a therapeutic agent that modulates the target of 2.
Drafting Endeavors
“But be careful what you wish for 'Cause you just might get it.”
Product-By-Process Claims
Process that Defines the Product
Purpose of product-by-process patent claims is to allow inventors to claim an otherwise patentable product that resists definition by other than the process by which it is made.
“Sometimes it’s damned hard to tell the dancer from the dance.”
Distinctive Structural Characteristics
The structure implied by the process steps of a product-by-process limitation should be considered especially (i) where the product can only be defined by the process steps, or (ii) where the manufacturing process steps would be expected to impart distinctive structural characteristics to the final product. See M.P.E.P. § 2113.
Validity v/s Infringement
The asymmetry regarding validity and infringement of a product-by-process claim is typically to the disadvantage of the patentee; claimed products can be anticipated by products made by a different process, but the claimed products are infringed only by those products made by the same process. See Amgen v. F. Hoffmann-La Roche, 580 F.3d 1340.
PrimateAI has many novelties like its non-human primate species training data, deep residual model architectures, bootstrapped training techniques, etc.
We select PrimateAI’s non-human primate training data as the source that imparts distinctive empirical structure to PrimateAI’s output, i.e., the PrimateAI Scores, and craft the following claim.
memory storing respective benign-eliminated pathogenicity scores for respective variants in a genome, wherein the benign-eliminated pathogenicity scores are generated by:
(a) eliminating from a pathogenic set of variants those common missense variants that are observed in non-human primates but are clinically benign in humans;
(b) including in a benign set of variants the common missense variants that are observed in the non-human primates and common missense variants that are observed in humans;
(c) assigning a pathogenic label to first protein sequences containing the pathogenic set, and a benign label to second protein sequences containing the benign set; and
(d) training a classifier to process the the first and second protein sequences as inputs, and to generate the benign-eliminated pathogenicity scores as outputs.
A system, comprising:
Our p-b-p claim uses the phrase “benign-eliminated pathogenicity scores” to cover the PrimateAI Scores as a novel product.
Steps (a) to (d) form a novel process without the AI intricacies.
The claim language is comprehensively supported by PrimateAI’s patent disclosure.
Dependent claims can add further novelty to the “benign-eliminated pathogenicity scores” by using other novel features of PrimateAI like “protein structure-based,” “evolutionary conversation-based,” and “ensembled.”
Product Claims
“Let’s dance.”
Having claims characterized as “product-by-process” claims renders them more vulnerable to patentability/validity attacks. Similarly, the scope of enforceability afforded by product-by-process claims may be subject to extremely narrow interpretation, as the alleged infringer may escape if using a different, non-equivalent process.
Therefore, we now craft “product claims” based on the novel product that came oozing out off our “product-by-process” claims.
Our product claim on the PrimateAI Scores is inspired by Enfish and McRo line of cases that conferred patentability on novel data structures which improved the accuracy and effect provided by a computer.
A computer system performing pathogenicity analysis, comprising:
memory configured with a pathogenicity table, the pathogenicity table including respective benign-eliminated pathogenicity scores for respective variants in a genome, wherein the benign-eliminated pathogenicity scores improve accuracy of the pathogenicity analysis by identifying as benign those common missense variants that are observed in non-human primates but are clinically benign in humans; and
means for retrieving the benign-eliminated pathogenicity scores in the pathogenicity table to perform the pathogenicity analysis with the improved accuracy.
Our product claim is now ripe for use in our downstream claims.