Take2 Tech Ltd v. Pacific Biosciences Of California Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Take2 Technologies Limited (Cayman Islands); The Chinese University of Hong Kong (Hong Kong)
  • Defendant: Pacific Biosciences of California, Inc. (Delaware)
  • Plaintiff's Counsel: Young Conaway Stargatt & Taylor, LLP
• Case Identification: 1:22-cv-01595, D. Del., 12/14/2022
• Venue Allegations: Venue is asserted based on Defendant's incorporation in Delaware and its transaction of business within the district.
• Core Dispute: Plaintiff alleges that Defendant's DNA sequencing platforms and associated software infringe a patent related to methods for detecting epigenetic DNA modifications.
• Technical Context: The technology involves using kinetic data from real-time DNA sequencing to identify base modifications like methylation, a key marker in the field of epigenetics relevant to cancer and disease diagnostics.
• Key Procedural History: The complaint alleges a history of collaboration discussions between the parties, beginning after Plaintiffs published their "holistic kinetic" (HK) model in a scientific journal in January 2021. Plaintiffs allege they disclosed confidential details of the technology to Defendant under a non-disclosure agreement and notified Defendant of the patent's allowance and issuance. The complaint alleges Defendant subsequently ceased communications and commercialized the technology in its SMRT Link v11.0 software, released in April 2022.

Case Timeline

Date	Event
2019-08-16	'794 Patent Priority Date
2021-01-25	PNAS article on HK model published
2021-02-19	First meeting between Plaintiffs and Defendant
2021-03-11	Second meeting between Plaintiffs and Defendant
2021-04-11	Plaintiffs allegedly inform Defendant of '794 Patent allowance
2021-08-17	'794 Patent Issued
2021-10-18	Defendant's workshop at ASHG Annual Meeting
2022-04-20	Defendant releases SMRT Link v11.0 software
2022-12-14	Complaint Filing Date
2023-03-03	Revised Public Version of Complaint Filed

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 11,091,794 - Determination of Base Modifications of Nucleic Acids

Issued August 17, 2021

The Invention Explained

• Problem Addressed: The patent describes a need for more accurate techniques for measuring DNA methylation, such as 5-methylcytosine (5mC) Compl. ¶13 It notes that prior methods like bisulfite sequencing can degrade DNA and introduce biases, limiting their accuracy and ability to analyze long DNA molecules '794 Patent, col. 2:6-33
• The Patented Solution: The invention is a method for detecting base modifications directly from single-molecule, real-time (SMRT) sequencing data, eliminating the need for chemical pre-treatment like bisulfite conversion '794 Patent, col. 2:35-44 It employs a machine learning model that analyzes kinetic features-specifically the interpulse duration (IPD) and pulse width (PW)-in conjunction with the sequence context of nucleotides within a defined "measurement window" to determine the methylation status of a target base '794 Patent, abstract '794 Patent, FIG. 9
• Technical Importance: This approach allows for the direct, high-accuracy detection of epigenetic markers on long, native DNA strands, which is critical for studying the role of epigenetics in disease development and diagnosis Compl. ¶13

Key Claims at a Glance

• The complaint asserts at least independent claim 1 Compl. ¶36
• The essential elements of independent claim 1 are:
  • (a) Receiving data from measured optical pulses and obtaining values for nucleotide identity, position, pulse width, and interpulse duration.
  • (b) Creating an input data structure from a "window" of nucleotides, which includes properties like identity, position relative to a target, pulse width, and interpulse duration for each nucleotide in the window.
  • (c) Inputting this data structure into a model that has been trained on data structures with known modification states to recognize whether a modification is present.
  • (d) Using the model to determine if the modification is present in a nucleotide at the target position.
• The complaint does not explicitly reserve the right to assert dependent claims, but this is standard practice.

III. The Accused Instrumentality

Product Identification

• The accused instrumentalities are Defendant's Sequel® II, Sequel IIe, and Revio™ DNA sequencing systems when equipped with or used with SMRT® Link software v11.0, v11.1, or later (collectively, "the PacBio Products") Compl. ¶35

Functionality and Market Context

• The PacBio Products are alleged to perform "HiFi sequencing" that "detects modifications in native DNA" by analyzing the "kinetics of base incorporation" Compl. ¶38 This functionality is specifically enabled by the "5mC CpG Detection" feature in the SMRT Link v11.0 software, which uses a convolutional neural network (CNN) model to process polymerase kinetics and determine the methylation status of CpG sites Compl. ¶¶27, 62 The complaint provides a screenshot from a PacBio presentation illustrating that its HiFi sequencing "detects modifications in native DNA through impacts on the kinetics of base incorporation" Compl. ¶38, Fig. 3 The complaint also includes a diagram from PacBio's materials showing how the system uses fluorescence (to identify the base) and kinetics (to identify modifications) Compl. ¶41, Fig. 4 Another visual from a PacBio presentation shows a "Neural network to call CpG methylation in HiFi reads," which takes a "feature vector" as input to a TensorFlow CNN Compl. ¶50, Fig. 5

IV. Analysis of Infringement Allegations

11,091,794 Infringement Allegations

Claim Element (from Independent Claim 1)	Alleged Infringing Functionality	Complaint Citation	Patent Citation
(a) receiving data acquired by measuring pulses in an optical signal corresponding to nucleotides sequenced in a sample nucleic acid molecule and obtaining, from the data, values for the following properties: for each nucleotide: an identity of the nucleotide, a position of the nucleotide within the sample nucleic acid molecule, a width of the pulse corresponding to the nucleotide, and an interpulse duration representing a time between the pulse corresponding to the nucleotide and a pulse corresponding to a neighboring nucleotide;	The PacBio Products receive sequencing data from HiFi sequencing technology, which measures fluorescent pulses and kinetics. This provides the identity of the nucleotide (from the color of the pulse), its position (by alignment to a reference), the pulse width (time to incorporate a base), and the interpulse duration (time between incorporations).	¶¶41-45	col. 10:27-41
(b) creating an input data structure, the input data structure comprising a window of the nucleotides sequenced in the sample nucleic acid molecule, wherein the input data structure includes, for each nucleotide within the window, the properties: the identity of the nucleotide, a position of the nucleotide with respect to a target position within the window, the width of the pulse corresponding to the nucleotide, and the interpulse duration;	The PacBio Products are alleged to perform a "windowing" process that creates a "feature vector" (e.g., a 16-base or 22-bp window around a CpG site). This feature vector, derived from a BAM file, is alleged to contain the nucleotide identity, position, pulse width, and interpulse duration for the nucleotides within the window.	¶¶48-53	col. 11:4-13
(c) inputting the input data structure into a model, the model trained by: receiving a first plurality of first data structures... wherein the modification has a known first state... storing a plurality of first training samples... and optimizing... parameters of the model based on outputs of the model matching or not matching corresponding labels...	The PacBio Products' "primrose" tool allegedly inputs the feature vector into a convolutional neural network (CNN). The complaint alleges this CNN is trained using "in vitro modified controls of methylated and unmethylated human DNA," which correspond to the claimed training samples with known modification states, and that the model is optimized based on these samples.	¶¶56-60	col. 11:44-67
(d) determining, using the model, whether the modification is present in a nucleotide at the target position within the window in the input data structure.	The output of the CNN model is alleged to be a "probability scale measure of whether the CpG is symmetrically 5mC-modified," which is used to determine if the modification is present at the target position. This determination is output in standard BAM format tags.	¶¶62-63	col. 11:6-9

• Identified Points of Contention:
  • Scope Questions: A potential point of contention is whether the "feature vector" created by the accused products from a BAM file Compl. ¶¶50-51 constitutes an "input data structure comprising a window of the nucleotides sequenced" as recited in claim 1(b). A court may need to determine if this claim language reads on a data structure that has been processed and formatted (into a BAM file and then a feature vector) rather than being the direct raw data from the sequenced window.
  • Technical Questions: The infringement case may turn on whether PacBio's alleged training process for its CNN model Compl. ¶59 meets all the specific sub-steps of the "model trained by" limitation in claim 1(c). Evidence will be required to establish that PacBio's process involves "receiving" data structures, "storing" training samples with labels, and "optimizing" parameters based on matching outputs to those labels, as specifically required by the claim language.

V. Key Claim Terms for Construction

• The Term: "input data structure"

  • Context and Importance: This term is central to the infringement theory. Claim 1(b) defines it as "comprising a window of the nucleotides sequenced" and including specific properties (identity, position, pulse width, interpulse duration). The complaint alleges that PacBio's "HiFi reads and feature vectors created from the BAM file" meet this limitation Compl. ¶51 The dispute may focus on whether this processed and formatted "feature vector" is equivalent to the claimed "input data structure."
  • Intrinsic Evidence for Interpretation:
    • Evidence for a Broader Interpretation: The patent's description of compiling kinetic features and sequence context into a "2-dimensional (i.e., 2-D) matrix as a measurement window" could support an interpretation that the "input data structure" is a formatted representation of the sequencing data, not necessarily the raw data itself '794 Patent, col. 23:20-25
    • Evidence for a Narrower Interpretation: The plain language "comprising a window of the nucleotides sequenced" could be argued to require a more direct representation of the sequence data from the window, rather than a derived "feature vector" that has undergone intermediate processing and formatting into a BAM file. The patent's figures, such as FIG. 4, depict the measurement window as a direct view of the sequenced DNA strand '794 Patent, Fig. 4

• The Term: "model trained by"

  • Context and Importance: Claim 1(c) requires inputting the data into a "model, the model trained by" a specific, multi-step process. Infringement hinges on whether PacBio's "primrose" CNN model Compl. ¶56 is trained using the recited method. Practitioners may focus on this term because the method of training is a specific functional limitation on the claimed model itself.
  • Intrinsic Evidence for Interpretation:
    • Evidence for a Broader Interpretation: The patent specification describes training statistical models using reference patterns from samples with known modification states, which is a general description that could encompass PacBio's alleged use of methylated and unmethylated control DNA '794 Patent, col. 26:30-34 '794 Patent, FIG. 9
    • Evidence for a Narrower Interpretation: The claim recites a specific sequence of training steps: "receiving a first plurality of first data structures," "storing a plurality of first training samples," and "optimizing... parameters." An argument could be made that this requires a specific training architecture and data flow, and any deviation by PacBio, or the use of a pre-trained model not trained in this exact manner, would fall outside the claim scope.

VI. Other Allegations

• Indirect Infringement: The complaint alleges that PacBio actively induces infringement by providing customers with user guides, reference guides for its SMRT® Link and SMRT® Tools software, and other resources that instruct them on how to use the products to perform the patented method of detecting 5mC modifications Compl. ¶66
• Willful Infringement: Willfulness is alleged based on both pre-suit and post-suit knowledge. The complaint alleges PacBio knew of the patent's priority application as early as March 2021, was informed of the patent's allowance in April 2021, and knew of the patent's issuance in August 2021, all prior to the April 2022 release of the accused SMRT Link v11.0 software Compl. ¶¶19, 21, 27, 67 It further alleges Plaintiffs' representative explicitly informed PacBio of its infringement in May 2022 Compl. ¶67

VII. Analyst's Conclusion: Key Questions for the Case

• A core issue will be one of definitional scope: Can the term "input data structure", defined in the patent as "comprising a window of the nucleotides sequenced," be construed to cover the accused "feature vector," which is a processed data set derived from an industry-standard BAM file format rather than the raw sequencing data itself?
• A second central question will be one of operational equivalence: Does the training process for Defendant's "primrose" CNN, which allegedly uses in-vitro modified DNA controls, satisfy the specific, sequential limitations of "receiving," "storing," and "optimizing" as recited in the "model trained by" clause of Claim 1, or is there a material difference in the training methodology?