PRIVILEGED AND CONFIDENTIAL — ATTORNEY WORK PRODUCT
This memorandum is protected by the attorney-client privilege and the work product doctrine. It has been prepared by counsel at Calderwood & Harkness LLP in anticipation of litigation and at the direction of the client, Genome Dx Therapeutics Inc. This memorandum and its contents should not be disclosed to any third party without the express written consent of Calderwood & Harkness LLP and Genome Dx Therapeutics Inc. Any unauthorized disclosure may result in waiver of applicable privileges.
MEMORANDUM
TO: Catherine Bloom, Responsible Partner, intellectual Property, Calderwood & Harkness LLP
FROM: Peter Vassiliev, Lead Associate, intellectual Property; Wendy Schultz, Working Associate, intellectual Property
DATE: March 10, 2025
RE: Pre-Filing Investigation — Genome Dx Therapeutics Inc. v. Helios Precision Diagnostics, Inc. — Analysis of Patent Infringement Claims Under U.S. Patent Nos. 10,438,271; 11,204,893; and 11,891,647
MATTER: GenomeDx / Helios Patent Litigation (Pre-Filing)
Right-click to update Table of Contents
I. EXECUTIVE SUMMARY
Calderwood & Harkness LLP has been engaged by Genome Dx Therapeutics Inc. ("GenomeDx" or "GDT"), a Delaware corporation headquartered at 4750 Torrey Pines Road, Suite 300, San Diego, California 92130, to evaluate the merits of filing a patent infringement action against Helios Precision Diagnostics, Inc. ("Helios"), a Delaware corporation headquartered at 200 CambridgePark Drive, Suite 600, Cambridge, Massachusetts 02140. The proposed action arises from Helios's development, commercialization, marketing, and sale of its ProstaSignal™ Genomic Classifier test, which GenomeDx believes infringes multiple claims of three United States patents owned by GenomeDx.
GenomeDx is a molecular diagnostics company principally engaged in the development and commercialization of genomic classifier tests for cancer prognosis and treatment guidance. The company's majority equity holder is Pinnacle Biosciences Capital, LP, which holds a 62.4% equity interest. For the fiscal year ending December 31, 2024, GenomeDx reported revenue of $387.6 million and earnings before interest, taxes, depreciation, and amortization ("EBITDA") of $54.3 million, representing a 14.0% margin. GenomeDx's primary revenue-generating product is the Decipher® Prostate genomic classifier test, which is widely recognized as the leading genomic prognostic test for prostate cancer in the post-radical prostatectomy clinical setting.
Helios launched its ProstaSignal™ test commercially on October 7, 2024. Based on our investigation, between October 2024 and March 2025, Helios has performed approximately 8,400 ProstaSignal™ tests at an average reimbursement rate of $3,820 per test, generating estimated revenue of $32,088,000. This commercial activity has directly impacted GenomeDx's market share, which has declined from approximately 74% in the genomic prostate cancer prognostic testing market prior to the ProstaSignal™ launch to approximately 68% as of the date of this memorandum.
GenomeDx holds a robust intellectual property portfolio consisting of 47 issued United States patents, 12 pending United States patent applications, and 19 international family members across key jurisdictions. Following a comprehensive review of this portfolio in light of the ProstaSignal™ product, this memorandum evaluates the assertion of three patents that are most directly implicated by the accused product:
• U.S. Patent No. 10,438,271 (the "'271 Patent") — "Methods for Predicting Prostate Cancer Outcomes Using Multi-Gene Expression Classifiers"
• U.S. Patent No. 11,204,893 (the "'893 Patent") — "Genomic Classifier Systems and Methods for Post-Surgical Cancer Prognosis"
• U.S. Patent No. 11,891,647 (the "'647 Patent") — "Improved Methods for Genomic Risk Stratification in Prostate Cancer Using Combined Biomarker and Clinical Variable Models"
Our analysis yields the following summary conclusions:
1. Infringement. GenomeDx has strong literal infringement claims against Helios under the '271 and '647 Patents. Each element of the independent claims of these patents is satisfied by the ProstaSignal™ product as currently marketed and sold. The '893 Patent presents a literal infringement question with respect to the "random-forest classifier" limitation in independent Claim 1, as ProstaSignal™ employs a gradient-boosted decision tree classifier rather than a random-forest classifier. However, a viable theory of infringement under the doctrine of equivalents exists for the '893 Patent, subject to a moderate risk of prosecution history estoppel.
2. Validity. Validity risks exist under 35 U.S.C. §§ 101, 102, and 103, but each is manageable based on our review of the prosecution histories, the prior art landscape, and current Federal Circuit jurisprudence. The '647 Patent presents the strongest validity position, supported by a 37 C.F.R. § 1.132 declaration establishing unexpected results and by claim language that integrates clinical covariates with genomic data.
3. Enforceability. No enforceability concerns have been identified. Our review of the prosecution files reveals no inequitable conduct, patent misuse, or other enforceability defenses available to Helios.
4. Forum. We recommend filing in the U.S. District Court for the Northern District of California, where venue and personal jurisdiction can be established based on Helios's California operations, marketing activities, and sales within the district.
5. Damages. Preliminary damages estimates range from approximately $5.8 million under a reasonable royalty theory to approximately $18.5 million under a lost profits theory for the infringement period to date (October 2024 through March 2025), with ongoing accrual at an estimated rate of $3 million to $5 million per month. Enhanced damages may be available if willful infringement is established.
Based on these conclusions, we recommend that GenomeDx proceed promptly with the filing of a complaint for patent infringement. The detailed analysis supporting this recommendation is set forth below.
II. BACKGROUND
A. The Client — Genome Dx Therapeutics Inc.
Genome Dx Therapeutics Inc. was founded in 2011 and is organized as a corporation under the laws of the State of Delaware. GenomeDx is qualified to do business in the State of California, where its principal offices are located at 4750 Torrey Pines Road, Suite 300, San Diego, California 92130.
1. Key Management
GenomeDx is led by the following members of senior management, each of whom may be relevant to the anticipated litigation as fact witnesses, inventors, or corporate representatives:
• Chief Executive Officer: Dr. Miriam Langford, PhD. Dr. Langford holds a doctorate in molecular biology from Stanford University and has served as CEO since 2017. She provides overall strategic direction for the company and has been instrumental in the commercialization of the Decipher® platform.
• Chief Technology Officer: Dr. Rajan Mehta, PhD. Dr. Mehta holds a doctorate in computational genomics from the Massachusetts Institute of Technology. He is a named inventor on all three patents-in-suit and has been the principal architect of GenomeDx's genomic classifier technology from its inception. Dr. Mehta will be a critical fact and potentially expert witness in the anticipated litigation.
• General Counsel: Victoria Sung, JD. Ms. Sung is a graduate of Georgetown University Law Center and oversees GenomeDx's legal affairs, including intellectual property strategy, regulatory compliance, and corporate governance. Ms. Sung has been the primary client contact for this engagement.
• Vice President, intellectual Property: Dr. Kevin Ostrowski, PhD, JD. Dr. Ostrowski holds a doctorate in molecular biology and a juris doctor degree, and he manages GenomeDx's patent prosecution, portfolio strategy, and IP licensing programs. Dr. Ostrowski is a named inventor on the '647 Patent.
• Chief Financial Officer: Thomas Brennan, MBA. Mr. Brennan holds an MBA from the Wharton School of the University of Pennsylvania and is responsible for GenomeDx's financial operations, including revenue recognition, budgeting, and investor relations. Mr. Brennan will be relevant to damages analysis and financial disclosures.
GenomeDx currently employs approximately 820 individuals, including approximately 340 laboratory and research-and-development staff located primarily at its San Diego headquarters and adjacent laboratory facilities.
2. Ownership Structure
GenomeDx's equity ownership is structured as follows:
Pinnacle Biosciences Capital, LP is a Delaware limited partnership managed by its general partner, Pinnacle Biosciences Capital GP, LLC. Pinnacle's Managing Partner is Jonathan Hargrove. The fund manages approximately $4.2 billion in assets under management ("AUM") across three investment vehicles. Pinnacle's principal office is located at 600 Montgomery Street, Suite 2200, San Francisco, California 94111. Pinnacle holds three of seven seats on GenomeDx's board of directors. While Pinnacle is the majority equity holder, it does not hold any direct interest in GenomeDx's patents, which are solely owned by GenomeDx as the corporate entity. This is relevant to the standing analysis set forth in Section VI.D below.
Ridgeline Health Ventures, a Series B investor holding 7.0% equity, is a healthcare-focused venture capital fund based in Menlo Park, California. Ridgeline holds one board seat and has been a supportive investor in GenomeDx's growth strategy.
3. Business Overview and Products
GenomeDx's primary product is the Decipher® Prostate genomic classifier test, a 22-gene expression-based prognostic test designed for patients who have undergone radical prostatectomy for prostate cancer. The Decipher® test measures the expression levels of 22 specific genes in tumor tissue obtained from formalin-fixed paraffin-embedded ("FFPE") surgical specimens, applies a trained random-forest machine-learning classifier to those expression values, and generates a genomic risk score (the "Decipher Score") that predicts the likelihood of clinical recurrence, metastasis, and prostate cancer-specific mortality within defined time horizons. Patients are classified into Low, Intermediate, and High genomic risk categories based on the Decipher Score, and this classification is used by urologists and oncologists to guide adjuvant treatment decisions.
GenomeDx holds the Multianalyte Assay with Algorithmic Analysis ("MAAA") code 0047M for the Decipher® test, which was obtained in 2018 from the American Medical Association ("AMA") Current Procedural Terminology ("CPT") Editorial Panel. This MAAA code is the basis for billing and reimbursement through Medicare and commercial insurance carriers. The Decipher® test is reimbursed by Medicare under the Molecular Diagnostic Services ("MolDX") program administered by Palmetto GBA.
The Decipher® test is currently marketed and sold as a laboratory-developed test ("LDT") under the Clinical Laboratory Improvement Amendments ("CLIA") regulatory framework. GenomeDx submitted a De Novo classification application to the U.S. Food and Drug Administration ("FDA") on August 15, 2024, seeking formal FDA authorization for the Decipher® test as a Class II medical device. That application remains under review and is not expected to materially affect the anticipated litigation.
4. Financial Profile
For the fiscal year ending December 31, 2024, GenomeDx reported the following key financial metrics:
GenomeDx's auditor is Whitfield & Garner LLP, located in San Diego, California. GenomeDx's primary lending relationship is with Meridian Commercial Bank, N.A., which provides a $75 million revolving credit facility. GenomeDx's registered agent in Delaware is Pacific Registered Agents, Inc., located at 1209 Orange Street, Wilmington, Delaware 19801.
5. Market Position
Prior to the commercial launch of the ProstaSignal™ test in October 2024, GenomeDx held approximately 74% of the U.S. market for genomic prostate cancer prognostic testing in the post-radical prostatectomy setting. This market share has declined to approximately 68% following the ProstaSignal™ launch. The decline in market share is directly attributable to the ProstaSignal™ test, which competes for the same patient population, clinical indication, ordering physician base, and reimbursement pathway as the Decipher® test. This competitive overlap is central to the lost profits damages analysis and the injunctive relief analysis presented in this memorandum.
6. IP Enforcement Insurance
GenomeDx carries intellectual property enforcement insurance through Greystone Specialty Insurance Co. under policy number GSI-IP-2024-07812. The policy provides $10 million in coverage for offensive patent litigation costs, subject to a $2.5 million deductible and an 80/20 co-insurance structure above the deductible up to the policy limit. The policy period runs from January 1, 2025, through December 31, 2025, with a renewal pending. Notice of potential claim should be provided to Greystone promptly upon the decision to proceed with filing, in accordance with the policy's notice provisions. We have reviewed the policy terms and do not anticipate coverage disputes with respect to this matter, though Greystone's consent requirements for outside counsel selection and litigation budgets should be confirmed.
B. The Accused Infringer — Helios Precision Diagnostics, Inc.
Helios Precision Diagnostics, Inc. was founded in 2019 and is organized as a corporation under the laws of the State of Delaware. Helios is qualified to do business in the Commonwealth of Massachusetts, where its corporate headquarters are located at 200 CambridgePark Drive, Suite 600, Cambridge, Massachusetts 02140, and in the State of California, where it operates a CLIA-certified diagnostic laboratory at 9830 Mesa Rim Road, Suite 140, San Diego, California 92121.
1. Key Management
Based on publicly available information, Helios's senior management team includes the following individuals:
• Chief Executive Officer: Dr. Nathan Ashford, MD, PhD. Dr. Ashford holds a medical degree and a doctorate in oncology and bioinformatics from Johns Hopkins University. He co-founded Helios in 2019 and has led the company's strategy to develop and commercialize genomic diagnostic products for urological cancers.
• Chief Scientific Officer: Dr. Elaine Kossack, PhD. Dr. Kossack holds a doctorate in genomics and is an alumna of the Broad Institute of MIT and Harvard. She has overseen the scientific development of the ProstaSignal™ test, including gene panel selection, classifier training, and clinical validation.
• Vice President, Commercial Operations: Derek Mallinson, MBA. Mr. Mallinson leads Helios's sales and marketing efforts, including the commercial launch of ProstaSignal™ and ongoing market development activities.
• General Counsel: Sara Whitmore, JD. Ms. Whitmore is a graduate of Harvard Law School and oversees Helios's legal affairs. Ms. Whitmore will likely serve as Helios's initial point of contact for litigation-related matters and will coordinate Helios's defense strategy with outside counsel.
Helios currently employs approximately 215 individuals, a substantial portion of whom are based at its Cambridge headquarters and San Diego laboratory.
2. Ownership Structure
Helios's equity ownership is structured as follows:
Corinth Ventures Fund III, LP, Helios's majority investor, is a venture capital fund managed by Priya Chandrasekaran with approximately $2.1 billion in AUM. Corinth's principal office is located at 399 Boylston Street, Suite 1100, Boston, Massachusetts 02116. Corinth's investment in Helios and its board representation may be relevant to discovery regarding Helios's awareness of GenomeDx's patent portfolio and any freedom-to-operate analyses conducted in connection with the ProstaSignal™ development program.
3. Financial Overview
For the fiscal year ending December 31, 2024, Helios reported revenue of $71.2 million. This figure includes revenue from the ProstaSignal™ test (launched in the fourth quarter of 2024) as well as revenue from other diagnostic products and services in Helios's portfolio.
4. Anticipated Outside Counsel
Based on our investigation, Helios is likely to retain Ridgecrest Durneval LLP as outside litigation counsel. Ridgecrest Durneval is a prominent intellectual property litigation firm with a strong track record in pharmaceutical and life sciences patent disputes. The anticipated lead partner is Marcus Pellegrini, who has more than 25 years of experience in patent litigation, including extensive experience defending diagnostic and genomic patent cases. Ridgecrest Durneval maintains offices at One Federal Street, Suite 3400, Boston, Massachusetts 02110, and at 350 South Grand Avenue, Suite 2800, Los Angeles, California 90071. Mr. Pellegrini is based in the Boston office but has practiced extensively in California federal courts. We expect a vigorous defense and should prepare accordingly.
C. The Accused Product — ProstaSignal™ Genomic Classifier
The ProstaSignal™ Genomic Classifier is a 24-gene expression-based prognostic test designed for prostate cancer patients who have undergone radical prostatectomy. The test is marketed by Helios as a tool to assist urologists and oncologists in making adjuvant treatment decisions following surgery. The commercial launch date was October 7, 2024.
1. Regulatory and Reimbursement Status
Helios obtained MAAA code 0239M for the ProstaSignal™ test from the AMA CPT Editorial Panel in September 2024, enabling billing and reimbursement through Medicare and commercial payers. Like the Decipher® test, ProstaSignal™ is classified and marketed as a laboratory-developed test under the CLIA framework. Both the ProstaSignal™ and Decipher® tests are reimbursed by Medicare under the MolDX program, which means they compete directly for the same reimbursement pathway and clinical adoption by the same population of ordering physicians.
2. Laboratory Operations
ProstaSignal™ is performed exclusively at Helios's CLIA-certified laboratory located at 9830 Mesa Rim Road, Suite 140, San Diego, California 92121. Tissue specimens are sent to this laboratory by ordering physicians and hospitals nationwide. The fact that all ProstaSignal™ testing occurs at a single, centralized laboratory is relevant to both the direct infringement analysis (Helios itself performs every step of the claimed methods at its laboratory) and the venue and jurisdictional analysis (the laboratory is located in California).
3. Technical Description
Based on our review of publicly available materials, including Helios's marketing brochures, clinical white papers, the MAAA code filing documentation, and published abstracts, the ProstaSignal™ test operates as follows:
• Sample Input: ProstaSignal™ operates on formalin-fixed paraffin-embedded ("FFPE") prostate tumor tissue obtained from radical prostatectomy specimens. The test requires submission of an FFPE tissue block or unstained slides.
• RNA Extraction and Expression Profiling: At the Helios laboratory, RNA is extracted from the FFPE tissue sample, and the expression levels of 24 specific genes are measured using a proprietary microarray and quantitative PCR-based methodology.
• Normalization: Gene expression data is normalized using a panel of reference genes to control for variability in sample quality and processing.
• Classifier Application: The normalized gene expression values are input into a gradient-boosted decision tree classifier — a supervised machine-learning algorithm that has been trained on a proprietary clinical cohort. This is a critical distinction from GenomeDx's Decipher® test, which employs a random-forest classifier. Both gradient-boosted decision trees and random forests are ensemble machine-learning methods that aggregate multiple individual decision trees, but they differ in the methodology by which the constituent trees are constructed and combined. In a random forest, trees are built independently on bootstrapped subsets of the training data with randomized feature selection, and predictions are aggregated by averaging (bagging). In gradient boosting, trees are built sequentially, with each subsequent tree trained to correct the residual errors of the ensemble assembled to that point, and predictions are aggregated additively.
• Output: The classifier generates a numerical output called the "ProstaScore," which ranges from 0 to 100. Based on the ProstaScore, patients are classified into three risk categories: Low, Intermediate, or High.
• Composite Report: The ProstaScore is combined with two clinical covariates — Gleason grade group and pre-operative prostate-specific antigen ("PSA") level — to generate a composite prognostic report. This integrated report is transmitted to the ordering healthcare provider.
4. Test Volume and Revenue
Based on our investigation, including analysis of public filings, investor presentations, MolDX billing data obtained through public records requests, and industry estimates, we estimate that Helios performed approximately 8,400 ProstaSignal™ tests between October 2024 and March 2025, at an average reimbursement rate of approximately $3,820 per test. This yields estimated ProstaSignal™ revenue of approximately $32,088,000 (8,400 × $3,820) for the period. These figures are preliminary and will be refined through discovery.
5. Gene Panel Overlap
A critical factual element of the infringement analysis is the degree of overlap between the gene panels used in the Decipher® and ProstaSignal™ tests. The Decipher® test measures 22 genes. The ProstaSignal™ test measures 24 genes. Our analysis, set forth in detail in the gene panel comparison table attached as Appendix C, reveals that 19 of ProstaSignal™'s 24 genes are identical to genes in the Decipher® 22-gene panel. Furthermore, the specification of the '271 Patent discloses a broader "predefined gene set" of 45 candidate biomarker genes from which the 22-gene Decipher® panel was selected. An additional 3 of ProstaSignal™'s 24 genes fall within this broader 45-gene predefined set, for a total of 22 ProstaSignal™ genes that fall within the 45-gene predefined gene set disclosed in the '271 Patent specification. Only 2 of ProstaSignal™'s 24 genes are entirely outside the disclosed gene set. This extensive overlap is strongly probative of infringement and may also support an inference of copying for purposes of the non-obviousness analysis.
III. PATENTS-IN-SUIT
This section describes the three patents recommended for assertion against Helios, including their prosecution histories, priority chains, key claims, and relevant claim construction issues. Full texts of the patents and prosecution history summaries are attached as Appendices A and B, respectively.
A. U.S. Patent No. 10,438,271 (the "'271 Patent")
Title: "Methods for Predicting Prostate Cancer Outcomes Using Multi-Gene Expression Classifiers"
The '271 Patent is the earliest-filed patent in the family and serves as the priority document for the '893 Patent (which claims the benefit of the '271 Patent's filing date as a continuation). The '271 Patent contains 24 issued claims, including 4 independent claims (Claims 1, 12, 18, and 22) and 20 dependent claims.
1. Claims Recommended for Assertion
We recommend asserting Claims 1, 3, 5, 7, 12, and 18 of the '271 Patent against Helios. These claims were selected based on their alignment with the ProstaSignal™ product, the breadth of their claim language, and their defensibility against anticipated invalidity challenges.
2. Key Independent Claim (Claim 1)
Claim 1 of the '271 Patent recites:
A method for prognosing the likelihood of clinical recurrence following radical prostatectomy in a patient diagnosed with prostate cancer, the method comprising:
(a) obtaining a biological sample containing tumor RNA from the patient;
(b) measuring expression levels of a panel of at least 22 genes selected from a predefined gene set;
(c) applying a trained machine-learning classifier to the measured expression levels to generate a numerical risk score; and
(d) classifying the patient into a risk category based on the numerical risk score.
This claim is notable for its use of the broad term "trained machine-learning classifier," which is not limited to any specific type of machine-learning algorithm. This breadth is significant because it encompasses both random-forest classifiers (as used in the Decipher® test) and gradient-boosted decision tree classifiers (as used in ProstaSignal™), thereby avoiding the claim construction difficulty that arises under the '893 Patent.
The term "predefined gene set" is defined in the specification at column 8, lines 14–45, which identifies a set of 45 candidate biomarker genes selected through a genome-wide expression analysis of prostate tumor specimens. The specification states that the 22-gene panel used in the preferred embodiment was selected from this 45-gene set based on statistical significance and prognostic value. The claim language requires "at least 22 genes selected from" the predefined gene set, which encompasses any panel of 22 or more genes drawn from the disclosed 45-gene set.
3. Prosecution History
The '271 Patent was prosecuted through three office actions. During the first office action, the Examiner issued a rejection under 35 U.S.C. § 101, characterizing the claims as directed to an abstract idea (a mathematical correlation between gene expression and cancer outcomes). The applicant responded by amending the claims to recite specific technical steps — including RNA extraction from FFPE samples and reference-gene normalization — and by arguing that the claims are directed to a specific, concrete diagnostic method that transforms patient tissue into a clinically actionable prognostic classification. The Examiner withdrew the § 101 rejection in the second office action.
The Examiner also issued rejections under § 102 (based on a non-analogous reference that was subsequently withdrawn) and § 103 (combining references that did not teach the specific gene set or classifier methodology). These rejections were overcome through a combination of amendments and arguments. The applicant cited and distinguished Klein et al. and Thornberry et al. in Information Disclosure Statements filed during prosecution.
The '271 Patent issued on October 8, 2019, following a Notice of Allowance dated July 15, 2019.
4. Named Inventor Status
Dr. Rajan Mehta remains employed by GenomeDx as CTO and is fully available as a witness.
Dr. Carolyn Xu departed GenomeDx in 2023 and is currently employed by Cascadia Genomics, Inc., a Washington corporation. Dr. Xu's departure was amicable according to client representatives, but she is no longer under any contractual obligation to cooperate with GenomeDx in litigation matters beyond the general obligations of a named inventor. Her availability as a fact witness should be confirmed early in the litigation process, as discussed in Section IX.E below.
Dr. David Petrossian is a named inventor on the '271 Patent, but his current employment and contact information are unknown to our client. Efforts to locate Dr. Petrossian should be undertaken promptly, as his testimony may be relevant to the inventive process, the selection of the gene panel, and the prosecution history.
B. U.S. Patent No. 11,204,893 (the "'893 Patent")
Title: "Genomic Classifier Systems and Methods for Post-Surgical Cancer Prognosis"
The '893 Patent is a continuation of the '271 Patent and claims the benefit of the same June 14, 2016 filing date. It shares the same specification but contains claims directed to more specific aspects of the genomic classifier methodology, including the random-forest classifier type and the minimum training cohort size. The '893 Patent contains 22 issued claims, including 3 independent claims (Claims 1, 14, and 20) and 19 dependent claims.
1. Claims Recommended for Assertion
We recommend asserting Claims 1, 2, 9, 14, and 20 of the '893 Patent. While literal infringement of the "random-forest classifier" limitation is problematic, the inclusion of this patent in the assertion portfolio provides additional coverage and supports a doctrine-of-equivalents theory.
2. Key Independent Claim (Claim 1)
Claim 1 of the '893 Patent recites:
A method for determining a prognosis for a prostate cancer patient following radical prostatectomy, the method comprising:
(a) normalizing gene-expression data obtained from a formalin-fixed paraffin-embedded prostate tissue sample using a reference gene panel comprising at least three reference genes;
(b) inputting the normalized expression values for a panel of prognostic genes into a random-forest classifier trained on a cohort of at least 500 patient samples to generate a prognostic score; and
(c) outputting the prognostic score, wherein the prognostic score is a numerical value correlated with the risk of metastasis within 5 years of radical prostatectomy.
The most significant claim construction issue for this patent is the term "random-forest classifier." This term is used in independent Claims 1, 14, and 20 and is incorporated by reference into each of the dependent claims recommended for assertion. As discussed in detail in the infringement analysis at Section IV.B below, ProstaSignal™ uses a gradient-boosted decision tree classifier rather than a random-forest classifier, creating a literal infringement gap.
3. Prosecution History
The '893 Patent was prosecuted through two office actions. The Examiner rejected Claims 1–22 under 35 U.S.C. § 102(a)(1) based on Nakamura et al., U.S. Patent No. 9,617,340 ("Nakamura"), issued April 4, 2017. Nakamura discloses gene expression profiling methods for cancer prognosis but does not disclose a random-forest classifier trained on a cohort of at least 500 patient samples.
The applicant overcame the Nakamura rejection by amending independent Claims 1, 14, and 20 to add the limitation "random-forest classifier trained on a cohort of at least 500 patient samples" and by arguing that Nakamura does not disclose or suggest this specific classifier type or training methodology. The Examiner accepted this distinction and issued a Notice of Allowance.
This prosecution history is directly relevant to the doctrine-of-equivalents analysis. The applicant's amendment adding "random-forest classifier" to overcome the Nakamura rejection creates a potential prosecution history estoppel argument. The amendment narrowed the claims from a broader formulation to specifically recite "random-forest classifier," and the applicant relied on this narrowing to distinguish the prior art. Under Festo Corp. v. Fenmore & Co., 535 U.S. 722 (2002), there is a presumption that the patentee surrendered the territory between the original claim scope and the amended claim scope. GenomeDx would need to rebut this presumption under one of the three recognized Festo exceptions (unforeseeability, tangential relation, or other reason), as discussed in Section IV.B.2 below.
Nakamura was cited in an Information Disclosure Statement and fully considered by the Examiner, which eliminates any inequitable conduct concern regarding this reference.
4. Named Inventor Status
Dr. Rajan Mehta and Dr. Carolyn Xu are the named inventors, with the same status and availability considerations discussed above.
C. U.S. Patent No. 11,891,647 (the "'647 Patent")
Title: "Improved Methods for Genomic Risk Stratification in Prostate Cancer Using Combined Biomarker and Clinical Variable Models"
The '647 Patent is the most recently issued patent in the portfolio and has the latest expiration date (September 9, 2041 — approximately 16.5 years from the date of this memorandum). It is directed to methods that combine a genomic risk score with clinical covariates to generate a composite prognostic metric, reflecting GenomeDx's next-generation approach to integrated risk stratification. The '647 Patent contains 20 issued claims, including 3 independent claims (Claims 1, 11, and 15) and 17 dependent claims. A terminal disclaimer was filed over the '271 and '893 Patents during prosecution.
1. Claims Recommended for Assertion
We recommend asserting Claims 1, 4, 6, 11, and 15 of the '647 Patent. These claims are the strongest in GenomeDx's portfolio for assertion against ProstaSignal™ due to their breadth (use of the generic term "classifier" rather than "random-forest classifier") and their recitation of the combined genomic-clinical covariate methodology that ProstaSignal™ employs.
2. Key Independent Claim (Claim 1)
Claim 1 of the '647 Patent recites:
A method for generating an integrated prognostic report for a prostate cancer patient, the method comprising:
(a) performing RNA extraction and expression profiling on a prostate tumor specimen obtained from the patient following radical prostatectomy;
(b) computing a genomic risk score using a classifier comprising at least 20 gene features, wherein the classifier has been trained on a cohort of patient samples with known clinical outcomes;
(c) combining the genomic risk score with at least two clinical covariates selected from the group consisting of Gleason score, prostate-specific antigen (PSA) level, surgical margin status, and pathological stage to generate a composite risk metric; and
(d) generating an integrated prognostic report comprising the composite risk metric and transmitting the integrated prognostic report to a healthcare provider.
This claim is notable for several features that distinguish it from the '893 Patent claims and make it the most straightforward to assert:
• Generic Classifier Language: The claim recites "a classifier comprising at least 20 gene features" — not "a random-forest classifier." This broader formulation encompasses gradient-boosted decision tree classifiers, random-forest classifiers, and other machine-learning classifiers, directly addressing the literal infringement gap identified in the '893 Patent analysis.
• Combined Biomarker-Clinical Covariate Model: The claim requires combining the genomic risk score with "at least two clinical covariates selected from the group consisting of Gleason score, prostate-specific antigen (PSA) level, surgical margin status, and pathological stage." ProstaSignal™ expressly combines its ProstaScore with Gleason grade group and pre-operative PSA — two of the enumerated covariates — to generate a composite prognostic report. This limitation maps precisely onto ProstaSignal™'s marketed functionality.
• Integrated Prognostic Report: The claim requires generating and transmitting an integrated prognostic report, which is a core feature of the ProstaSignal™ product as marketed to ordering physicians.
3. Prosecution History
The '647 Patent was prosecuted through three office actions. The Examiner initially rejected the claims under 35 U.S.C. § 103, combining Nakamura with Gagnon (WO 2018/065413). The applicant responded with a combination of claim amendments and a declaration under 37 C.F.R. § 1.132 submitted by Dr. Rajan Mehta.
Dr. Mehta's declaration presented clinical validation data demonstrating that the combined genomic-clinical covariate model claimed in the '647 Patent achieved statistically significant improvements in prognostic accuracy compared to (i) genomic-only models and (ii) clinical-variable-only models. Specifically, the combined model achieved an area under the receiver operating characteristic curve ("AUC-ROC") of 0.89 for 5-year metastasis prediction, compared to 0.78 for the genomic-only model and 0.72 for the clinical-variable-only model. This unexpected improvement was not predicted by or suggested in any of the cited prior art references. The Examiner found this evidence persuasive and withdrew the § 103 rejection.
A terminal disclaimer was filed over the '271 and '893 Patents to address a double-patenting rejection, which is a routine procedural step that does not affect the validity or enforceability of the '647 Patent's claims.
4. Named Inventor Status
Dr. Rajan Mehta and Dr. Kevin Ostrowski are both currently employed by GenomeDx and are fully available as witnesses.
Dr. Priya Shah is a named inventor on the '647 Patent. Dr. Shah is a former Senior Research Scientist at GenomeDx who contributed to the development of the combined biomarker-clinical variable model. Client representatives indicate that Dr. Shah remains in the San Diego area and is available to cooperate in litigation. Her current contact information should be confirmed.
IV. INFRINGEMENT ANALYSIS
This section presents our element-by-element analysis of Helios's infringement of each patent-in-suit, addressing both literal infringement and, where applicable, infringement under the doctrine of equivalents. Our analysis is based on publicly available information regarding the ProstaSignal™ test, including marketing materials, clinical white papers, the MAAA code filing, published abstracts, and investor presentations. We note that certain aspects of Helios's specific laboratory protocols, classifier training methodology, and internal documentation will require confirmation through discovery.
A. Literal Infringement — '271 Patent
1. Claim 1 Element-by-Element Analysis
We have analyzed each element of independent Claim 1 of the '271 Patent against the ProstaSignal™ product. Our analysis is as follows:
Element (a): "obtaining a biological sample containing tumor RNA from the patient"
ProstaSignal™ operates on tissue samples obtained from radical prostatectomy specimens. Ordering physicians submit FFPE tissue blocks or unstained slides to Helios's CLIA-certified laboratory in San Diego, where RNA is extracted from the tumor tissue as the first step of the assay process. Helios's published laboratory requisition form and marketing materials confirm that the test requires a biological sample (prostate tumor tissue) from which tumor RNA is obtained. This element is clearly satisfied.
Element (b): "measuring expression levels of a panel of at least 22 genes selected from a predefined gene set"
This element requires analysis of two sub-issues: (i) whether ProstaSignal™ measures expression levels of "at least 22 genes," and (ii) whether those genes are "selected from a predefined gene set."
As to the first sub-issue, ProstaSignal™ uses a 24-gene panel. Twenty-four is greater than twenty-two. The numerical threshold is satisfied on its face.
As to the second sub-issue, the critical question is the construction of the term "predefined gene set." The specification of the '271 Patent, at column 8, lines 14–45, discloses a predefined gene set comprising 45 candidate biomarker genes identified through genome-wide expression analysis of prostate tumor specimens. The 22-gene Decipher® panel was selected from this 45-gene set as the optimal prognostic subset. The claim language — "at least 22 genes selected from a predefined gene set" — requires that the measured genes be drawn from this disclosed set.
Our gene panel comparison (Appendix C) reveals the following:
• 19 of ProstaSignal™'s 24 genes are identical to genes in the 22-gene Decipher® panel.
• An additional 3 of ProstaSignal™'s 24 genes, while not part of the 22-gene Decipher® panel, fall within the broader 45-gene predefined gene set disclosed in the '271 Patent specification.
• Therefore, 22 of ProstaSignal™'s 24 genes are "selected from" the 45-gene predefined gene set.
• Only 2 of ProstaSignal™'s 24 genes fall outside the predefined gene set entirely.
Because 22 of ProstaSignal™'s 24 genes are drawn from the predefined gene set, and the claim requires "at least 22 genes selected from a predefined gene set," this element is satisfied under literal infringement.
We note that Helios may argue that the "predefined gene set" must be limited to the 22-gene Decipher® panel rather than the broader 45-gene set, or that "selected from" requires that all measured genes be drawn from the predefined gene set (rather than permitting additional genes outside the set). Under either construction, the analysis becomes more complex. However, the specification clearly defines the predefined gene set as the 45-gene candidate set, and the claim language "at least 22 genes selected from" plainly permits measurement of additional genes beyond the minimum 22. We believe the broader construction is well supported by the intrinsic evidence. Further factual investigation during discovery regarding Helios's specific gene identities and selection rationale will strengthen this analysis.
Element (c): "applying a trained machine-learning classifier to the measured expression levels to generate a numerical risk score"
ProstaSignal™ applies a gradient-boosted decision tree classifier to the measured gene expression values. Gradient boosting is a supervised machine-learning technique, and a gradient-boosted decision tree classifier is indisputably a "trained machine-learning classifier" within the plain and ordinary meaning of that term as understood by a person of ordinary skill in the art ("POSITA"). The '271 Patent specification does not limit "trained machine-learning classifier" to any specific algorithm or methodology; it uses the term generically to encompass any machine-learning model that has been trained on labeled data to generate a predictive output. The specification discloses random forests as the preferred embodiment but also references "other ensemble methods" and "supervised learning approaches" as alternatives (column 12, lines 35–42).
The ProstaSignal™ classifier generates a numerical output — the ProstaScore — which ranges from 0 to 100. This is a "numerical risk score" within the meaning of the claim.
This element is satisfied.
Element (d): "classifying the patient into a risk category based on the numerical risk score"
ProstaSignal™ classifies patients into three risk categories — Low, Intermediate, and High — based on the ProstaScore. This is precisely the classification step recited in the claim. The specific number of categories (three) and their labels are not limited by the claim language, which requires only that the patient be classified "into a risk category based on the numerical risk score."
This element is satisfied.
2. Dependent Claims 3, 5, 7, 12, and 18
The dependent claims recommended for assertion add the following additional limitations:
• Claim 3 requires that the biological sample is an FFPE sample. ProstaSignal™ processes FFPE tissue. Satisfied.
• Claim 5 requires that the gene panel includes specific named genes from a defined subset. Preliminary analysis, based on the gene panel comparison in Appendix C, indicates that ProstaSignal™'s panel includes the requisite named genes. Confirmation of specific gene identities through discovery is warranted.
• Claim 7 requires a normalization step using reference genes prior to classifier application. ProstaSignal™ employs normalization using reference genes. Likely satisfied, subject to confirmation of specific normalization methodology.
• Claim 12 (independent) is directed to a system claim reciting similar elements with additional hardware/software components. ProstaSignal™ is implemented on a computer system at Helios's laboratory. Likely satisfied; specific system architecture to be confirmed in discovery.
• Claim 18 (independent) requires the output to include a probability estimate of metastasis within a defined time period. ProstaSignal™'s ProstaScore is correlated with metastasis risk, and marketing materials reference 5-year metastasis probability. Likely satisfied.
3. Assessment. The literal infringement case for the '271 Patent is strong. All elements of independent Claim 1 are met by ProstaSignal™, including the critical gene panel limitation (22 of 24 ProstaSignal™ genes fall within the predefined gene set). The use of the broad term "trained machine-learning classifier" in the '271 Patent — rather than the more specific "random-forest classifier" of the '893 Patent — eliminates the most significant literal infringement obstacle. The dependent claims appear to add limitations that ProstaSignal™ also satisfies, though some require factual confirmation through discovery.
B. Literal Infringement — '893 Patent
1. Claim 1 Element-by-Element Analysis
Element (a): "normalizing gene-expression data obtained from a formalin-fixed paraffin-embedded prostate tissue sample using a reference gene panel comprising at least three reference genes"
ProstaSignal™ processes FFPE prostate tissue samples and employs a normalization step using reference genes. Helios's published methodology references the use of a reference gene panel for normalization purposes. The specific number of reference genes used by Helios has not been confirmed from public sources; if Helios uses at least three reference genes (which is standard laboratory practice for FFPE-based gene expression assays), this element is satisfied. This is a factual question that should be confirmed early in discovery through interrogatories or a Rule 30(b)(6) deposition.
This element is likely satisfied, subject to factual confirmation.
Element (b): "inputting the normalized expression values for a panel of prognostic genes into a random-forest classifier trained on a cohort of at least 500 patient samples to generate a prognostic score"
This is the critical limitation for the '893 Patent and the primary source of infringement risk. ProstaSignal™ uses a gradient-boosted decision tree classifier, not a random-forest classifier.
Random forests and gradient-boosted decision trees are both ensemble machine-learning methods that aggregate multiple decision trees, but they are technically distinct algorithms:
• Random Forests use bootstrap aggregation ("bagging"). Multiple decision trees are independently constructed on random subsets of the training data with randomized feature selection at each split node. The individual trees' predictions are aggregated (typically by averaging for regression or majority voting for classification) to produce the final output. The key characteristic is that each tree is independent of the others.
• Gradient-Boosted Decision Trees use boosting. Trees are constructed sequentially, with each new tree trained to minimize the residual errors of the ensemble built up to that point. The individual trees' predictions are combined additively, with each tree making an incremental correction to the prior ensemble. The key characteristic is that each tree depends on the errors of the previous trees.
A POSITA — whom we define as a researcher with a PhD in computational biology, bioinformatics, or a related field, with at least three years of experience in applying machine-learning methods to genomic data — would recognize these as distinct algorithms with different training procedures, different bias-variance trade-off profiles, and different hyperparameter tuning requirements. While both are "ensemble tree methods," the term "random-forest classifier" has a well-understood specific meaning in the machine-learning literature that does not encompass gradient-boosted decision trees.
Accordingly, literal infringement of this element is not established under the plain and ordinary meaning of "random-forest classifier."
With respect to the "trained on a cohort of at least 500 patient samples" sub-limitation, Helios's published clinical validation white paper references a training cohort of 612 patients with known outcomes. This sub-limitation would likely be satisfied if established, though the "random-forest classifier" gap is the threshold issue.
Element (c): "outputting the prognostic score, wherein the prognostic score is a numerical value correlated with the risk of metastasis within 5 years of radical prostatectomy"
ProstaSignal™ generates a ProstaScore from 0 to 100, and Helios's marketing materials describe the ProstaScore as correlated with the risk of metastasis within defined time horizons, including 5-year metastasis risk. This element is likely satisfied.
2. Doctrine of Equivalents — '893 Patent
Because literal infringement of the "random-forest classifier" limitation is not established, GenomeDx should assert infringement under the doctrine of equivalents ("DOE") for the '893 Patent claims. Under the DOE, a product or process infringes a patent claim if it performs substantially the same function, in substantially the same way, to achieve substantially the same result as the claimed element. Warner-Jenkinson Co. v. Harborline Enterprises Davis Chem. Co., 520 U.S. 17, 40 (1997).