[sheet: Cover] Field
Literal Infringement: GenomeDx contends that a gradient-boosted decision tree ensemble is a type of or functionally equivalent to a random-forest classifier, as both are ensemble methods based on decision trees. This position requires claim construction of 'random-forest classifier.' Helios will likely dispute literal infringement of this element.
Doctrine of Equivalents: The ProstaSignal™ GBDT classifier performs substantially the same function (combining multiple decision trees to generate a prediction), in substantially the same way (ensemble aggregation of decision-tree outputs trained on labeled patient data), to achieve substantially the same result (a numerical prognostic score predicting metastasis risk) as the claimed random-forest classifier. The difference — bagging (random forest) vs. boosting (GBDT) — is an insubstantial difference in ensemble methodology.
Note re Prosecution History Estoppel: During prosecution of the '893 Patent, the applicant distinguished Nakamura et al. (U.S. Patent No. 9,617,340) by arguing the specific random-forest classifier training methodology and minimum cohort size of 500 patients. The amendment narrowing to 'random-forest classifier' was made to distinguish Nakamura's neural-network approach, not to surrender all ensemble tree-based methods. The argument-based estoppel analysis should be evaluated carefully — the distinction was drawn against neural networks specifically, not against other decision-tree ensemble methods like GBDT. A tangential-relation or other-reason exception to prosecution history estoppel may apply.
Literal Infringement: Depends on construction of 'random-forest classifier' as discussed in Claim 1, Element 1[b]. If 'random-forest classifier' is construed broadly to encompass decision-tree ensemble methods, the 800-tree GBDT meets the 'at least 500 individual decision trees' limitation.
Doctrine of Equivalents: Even if 'random-forest classifier' is construed narrowly, the GBDT ensemble with ~800 trees performs the same function (aggregating predictions from a large number of decision trees), in the same way (iterative tree construction on training data), to achieve the same result (a robust prognostic score). The number of trees (800) exceeds the claimed minimum (500).
However, further analysis is needed: GenomeDx's position is that the ProstaSignal™ panel, when including reference/normalization genes used in the normalization step (Element 1[a]), may include additional genes from the predefined set that bring the total to 22 or more. Helios's full gene list (including reference genes) must be confirmed from discovery documents.
| Alternative: If literal infringement cannot be established (i.e., fewer than 22 genes from the predefined set), DOE analysis would consider whether the 5 non-overlapping genes perform substantially the same function in substantially the same way to achieve substantially the same result as the genes they replace. | ProstaSignal™ gene panel list (from Helios technical documentation, AEO); GenomeDx '893 Patent specification Table 2 (predefined gene set); comparison analysis of 24 ProstaSignal™ genes vs. patent gene set; Helios SOPs identifying reference genes used in normalization | Critical limitation. The 19-of-24 gene overlap may be insufficient for literal infringement of this dependent claim requiring ≥22 genes from the predefined set. Discovery into Helios's complete gene list (including reference genes) is priority. If only 19 overlap, DOE theory needed. This claim may be weaker than Claims 1, 14, and 20. [sheet: Claim 14] Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes 14[preamble] | A system for generating a post-surgical prostate cancer prognosis comprising: | Literal | Helios operates an integrated laboratory and computational system at its CLIA-certified facility (9830 Mesa Rim Road, Suite 140, San Diego, CA 92121) that performs the ProstaSignal™ test, generating post-radical-prostatectomy prognostic results for prostate cancer patients. | Helios CLIA certification; laboratory facility information; ProstaSignal™ system architecture documentation (AEO) | ProstaSignal™ constitutes a 'system' for generating a post-surgical prostate cancer prognosis. 14[a] | a gene-expression measurement module configured to receive a formalin-fixed paraffin-embedded prostate tissue sample and generate expression-level data for a plurality of genes; | Literal | Helios's laboratory includes RNA extraction and gene-expression profiling instrumentation (microarray or RNA-seq platform) configured to process FFPE prostate tissue samples and output expression-level data for the 24 genes in the ProstaSignal™ panel plus reference genes. | Helios laboratory SOPs; equipment lists (produced documents); ProstaSignal™ test requisition form | Literal infringement: The laboratory instrumentation constitutes a 'gene-expression measurement module' that receives FFPE samples and generates expression data. 14[b] | a processor executing a random-forest classification algorithm trained on a cohort of at least 500 patient samples, the algorithm configured to receive normalized expression-level data and output a numerical prognostic score correlated with metastasis risk within 5 years; | Both (Literal disputed; DOE primary) | Helios's computational infrastructure includes one or more processors executing the ProstaSignal™ GBDT classification algorithm, trained on ~1,200 patient samples, which receives normalized expression data and outputs the ProstaScore (0–100) correlated with 5-year metastasis risk. |
| Same literal/DOE analysis as Claim 1, Element 1[b]: The 'random-forest classification algorithm' limitation is the central dispute. ProstaSignal™ uses gradient-boosted decision trees rather than random forests. DOE theory: function-way-result test is met; prosecution history estoppel analysis applies (see Claim 1 chart and Prosecution History Estoppel Analysis Memorandum). | Helios algorithm documentation (AEO); system architecture documents; same evidence as Claim 1[b] | Mirrors Claim 1[b] analysis. The 'processor' and 'system' framing does not change the core dispute over 'random-forest' vs. GBDT. 14[c] | and a reporting module configured to transmit the numerical prognostic score to a healthcare provider. | Literal | ProstaSignal™ generates a patient report containing the ProstaScore and risk classification (Low/Intermediate/High) and transmits it electronically (via secure portal and/or fax) to the ordering healthcare provider. Helios's reporting system constitutes a 'reporting module.' | Sample ProstaSignal™ patient reports; Helios reporting system documentation; provider portal screenshots | Literal infringement: The electronic reporting system directly meets this limitation. [sheet: Claim 20] Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes 20[dependency] | The system of claim 14, wherein the reporting module is further configured to generate a report comprising the numerical prognostic score, a risk category classification, and a graphical representation of the patient's metastasis risk relative to a reference population. | Literal | The ProstaSignal™ patient report includes: (1) the numerical ProstaScore (0–100); (2) a risk category classification (Low, Intermediate, or High); and (3) a graphical chart showing the patient's ProstaScore plotted against a distribution curve of a reference patient population, visually illustrating the patient's relative metastasis risk. |
| All three sub-elements of this dependent claim are met by the ProstaSignal™ report format. | Sample ProstaSignal™ patient reports (multiple examples produced in discovery); Helios marketing materials showing sample report layout; ProstaSignal™ product brochure | Strong literal infringement. The ProstaSignal™ report format closely mirrors the claim language, including the graphical representation requirement. |
[sheet: Cover]
Field | Value
Document Title | Claim Chart — U.S. Patent No. 11,204,893 ('893 Patent) vs. Helios ProstaSignal™ Genomic Classifier
Matter | Genome Dx Therapeutics Inc. v. Helios Precision Diagnostics, Inc.
Case No. | 3:25-cv-01847-JWL
Court | U.S. District Court, Northern District of California
Assigned Judge | Hon. Jessica W. Langston
Patent-in-Suit | U.S. Patent No. 11,204,893
Patent Title | Genomic Classifier Systems and Methods for Post-Surgical Cancer Prognosis
Filing Date | March 22, 2019 (continuation of '271 Patent family)
Issue Date | December 21, 2021
Expiration Date | June 14, 2036 (same priority chain as U.S. Patent No. 10,438,271)
Named Inventors | Dr. Rajan Mehta; Dr. Carolyn Xu
Asserted Claims | Claims 1, 2, 9, 14, and 20
Accused Product | Helios Precision Diagnostics, Inc. — ProstaSignal™ Genomic Classifier
Accused Product Description | 24-gene expression-based prognostic test for prostate cancer post-radical prostatectomy; uses gradient-boosted decision tree classifier; commercial launch October 7, 2024
Prepared By | Peter Vassiliev, Lead Associate, Calderwood & Harkness LLP; Wendy Schultz, Working Associate, Calderwood & Harkness LLP
Reviewed By | Catherine Bloom, Responsible Partner, Calderwood & Harkness LLP
Client | Genome Dx Therapeutics Inc.
Date Prepared | March 2025 (updated February 2026)
Confidentiality | ATTORNEY WORK PRODUCT — PRIVILEGED AND CONFIDENTIAL
Designation | HIGHLY CONFIDENTIAL — ATTORNEYS' EYES ONLY (subject to Stipulated Protective Order entered December 18, 2025)
[sheet: Claim 1]
Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes
1[preamble] | A method comprising: | Literal | Helios performs the ProstaSignal™ Genomic Classifier as a laboratory-developed test (LDT) at its CLIA-certified laboratory at 9830 Mesa Rim Road, Suite 140, San Diego, CA 92121. The test is a method for post-surgical prostate cancer prognosis. MAAA code 0239M was obtained in September 2024; commercial launch occurred October 7, 2024. Approximately 8,400 tests performed from October 2024 through March 2025. | ProstaSignal™ marketing materials; Helios website; CMS MolDX MAAA code registry (code 0239M); Helios commercial launch press release (Oct. 7, 2024); Helios revenue data showing ~8,400 tests at avg. $3,820/test reimbursement | The ProstaSignal™ test is a method performed by Helios. The preamble recites 'a method' which is met by the step-by-step diagnostic process Helios performs on patient specimens.
1[a] | normalizing gene-expression data from a formalin-fixed paraffin-embedded prostate tissue sample using a reference gene panel; | Literal | ProstaSignal™ receives formalin-fixed paraffin-embedded (FFPE) prostate tissue specimens from pathology laboratories. RNA is extracted from the FFPE sample and gene-expression levels are measured. ProstaSignal™ normalizes the raw expression data using a set of housekeeping/reference genes prior to classifier input, as described in Helios's published validation studies and internal SOPs. | Helios laboratory SOPs (produced under Protective Order, designated Highly Confidential — AEO); ProstaSignal™ clinical validation white paper; Helios test requisition form specifying FFPE specimen requirement; deposition testimony of Dr. Elaine Kossack (anticipated) and Dr. Anton Rykov (anticipated) | Literal infringement: ProstaSignal™ uses FFPE prostate tissue and applies reference-gene normalization, directly meeting this limitation. The normalization step using a reference gene panel is confirmed in Helios's published methodology.
1[b] | inputting normalized expression values into a random-forest classifier trained on a cohort of at least 500 patient samples; | Both (Literal analysis disputed; DOE analysis primary) | ProstaSignal™ inputs normalized expression values from its 24-gene panel into a gradient-boosted decision tree (GBDT) classifier. The GBDT classifier was trained on a cohort of approximately 1,200 patient samples per Helios's validation publications. The ProstaSignal™ classifier is not described as a 'random-forest' classifier but rather uses gradient-boosted decision trees.
Literal Infringement: GenomeDx contends that a gradient-boosted decision tree ensemble is a type of or functionally equivalent to a random-forest classifier, as both are ensemble methods based on decision trees. This position requires claim construction of 'random-forest classifier.' Helios will likely dispute literal infringement of this element.
Doctrine of Equivalents: The ProstaSignal™ GBDT classifier performs substantially the same function (combining multiple decision trees to generate a prediction), in substantially the same way (ensemble aggregation of decision-tree outputs trained on labeled patient data), to achieve substantially the same result (a numerical prognostic score predicting metastasis risk) as the claimed random-forest classifier. The difference — bagging (random forest) vs. boosting (GBDT) — is an insubstantial difference in ensemble methodology.
Note re Prosecution History Estoppel: During prosecution of the '893 Patent, the applicant distinguished Nakamura et al. (U.S. Patent No. 9,617,340) by arguing the specific random-forest classifier training methodology and minimum cohort size of 500 patients. The amendment narrowing to 'random-forest classifier' was made to distinguish Nakamura's neural-network approach, not to surrender all ensemble tree-based methods. The argument-based estoppel analysis should be evaluated carefully — the distinction was drawn against neural networks specifically, not against other decision-tree ensemble methods like GBDT. A tangential-relation or other-reason exception to prosecution history estoppel may apply. | Helios validation study publications; Helios algorithm documentation (AEO); ProstaSignal™ technical white paper describing GBDT architecture; Helios internal development records re: classifier selection (produced documents); '893 Patent prosecution history (Amendment dated August 2021 distinguishing Nakamura); technical expert analysis from Dr. Franklin Ogawa (proposed, Stanford, Computational Oncology) regarding equivalence of random-forest and GBDT ensemble methods | This is the key disputed limitation. The '893 Patent specifically claims a 'random-forest classifier' while ProstaSignal™ uses a gradient-boosted decision tree. Literal infringement turns on claim construction. DOE is the stronger theory but must survive prosecution history estoppel analysis. The training cohort size requirement (≥500) is met by Helios's ~1,200-patient training set. See separate Prosecution History Estoppel Analysis Memorandum for detailed treatment.
1[c] | and outputting a numerical prognostic score correlated with metastasis risk within 5 years. | Literal | ProstaSignal™ outputs a 'ProstaScore' — a numerical score ranging from 0 to 100 that is correlated with the risk of metastasis within 5 years following radical prostatectomy. Patients are classified as Low (0–33), Intermediate (34–66), or High (67–100) risk based on the ProstaScore. The score is reported to the ordering healthcare provider. | ProstaSignal™ sample patient report; Helios marketing materials; ProstaSignal™ clinical validation study data; Helios website product description | Literal infringement: ProstaScore is a numerical prognostic score (0–100) expressly correlated with 5-year metastasis risk, directly satisfying this limitation.
[sheet: Claim 2]
Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes
2[dependency] | The method of claim 1, wherein the random-forest classifier comprises at least 500 individual decision trees. | Both (Literal analysis disputed; DOE analysis primary) | ProstaSignal™ uses a gradient-boosted decision tree ensemble. Per Helios's technical documentation (AEO), the GBDT model contains approximately 800 boosting iterations (i.e., 800 individual decision trees in the ensemble).
Literal Infringement: Depends on construction of 'random-forest classifier' as discussed in Claim 1, Element 1[b]. If 'random-forest classifier' is construed broadly to encompass decision-tree ensemble methods, the 800-tree GBDT meets the 'at least 500 individual decision trees' limitation.
Doctrine of Equivalents: Even if 'random-forest classifier' is construed narrowly, the GBDT ensemble with ~800 trees performs the same function (aggregating predictions from a large number of decision trees), in the same way (iterative tree construction on training data), to achieve the same result (a robust prognostic score). The number of trees (800) exceeds the claimed minimum (500). | Helios algorithm architecture documentation (AEO); Helios internal technical reports re: model hyperparameters; anticipated expert testimony from Dr. Franklin Ogawa | Depends on Claim 1 infringement. The numerical limitation (≥500 trees) is met by Helios's ~800-tree ensemble. The dispute centers on whether 'random-forest' encompasses GBDT.
[sheet: Claim 9]
Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes
9[dependency] | The method of claim 1, wherein the normalized expression values are derived from a panel of at least 22 genes selected from the gene set consisting of [enumerated predefined gene list in the '893 Patent specification]. | Literal | ProstaSignal™ measures expression levels from a 24-gene panel. 19 of the 24 ProstaSignal™ genes overlap with GenomeDx's patented predefined gene set as enumerated in the '893 Patent specification. Since the claim requires 'at least 22 genes selected from' the predefined gene set, and only 19 ProstaSignal™ genes are drawn from that set, this element is not literally met.
However, further analysis is needed: GenomeDx's position is that the ProstaSignal™ panel, when including reference/normalization genes used in the normalization step (Element 1[a]), may include additional genes from the predefined set that bring the total to 22 or more. Helios's full gene list (including reference genes) must be confirmed from discovery documents.
Alternative: If literal infringement cannot be established (i.e., fewer than 22 genes from the predefined set), DOE analysis would consider whether the 5 non-overlapping genes perform substantially the same function in substantially the same way to achieve substantially the same result as the genes they replace. | ProstaSignal™ gene panel list (from Helios technical documentation, AEO); GenomeDx '893 Patent specification Table 2 (predefined gene set); comparison analysis of 24 ProstaSignal™ genes vs. patent gene set; Helios SOPs identifying reference genes used in normalization | Critical limitation. The 19-of-24 gene overlap may be insufficient for literal infringement of this dependent claim requiring ≥22 genes from the predefined set. Discovery into Helios's complete gene list (including reference genes) is priority. If only 19 overlap, DOE theory needed. This claim may be weaker than Claims 1, 14, and 20.
[sheet: Claim 14]
Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes
14[preamble] | A system for generating a post-surgical prostate cancer prognosis comprising: | Literal | Helios operates an integrated laboratory and computational system at its CLIA-certified facility (9830 Mesa Rim Road, Suite 140, San Diego, CA 92121) that performs the ProstaSignal™ test, generating post-radical-prostatectomy prognostic results for prostate cancer patients. | Helios CLIA certification; laboratory facility information; ProstaSignal™ system architecture documentation (AEO) | ProstaSignal™ constitutes a 'system' for generating a post-surgical prostate cancer prognosis.
14[a] | a gene-expression measurement module configured to receive a formalin-fixed paraffin-embedded prostate tissue sample and generate expression-level data for a plurality of genes; | Literal | Helios's laboratory includes RNA extraction and gene-expression profiling instrumentation (microarray or RNA-seq platform) configured to process FFPE prostate tissue samples and output expression-level data for the 24 genes in the ProstaSignal™ panel plus reference genes. | Helios laboratory SOPs; equipment lists (produced documents); ProstaSignal™ test requisition form | Literal infringement: The laboratory instrumentation constitutes a 'gene-expression measurement module' that receives FFPE samples and generates expression data.
14[b] | a processor executing a random-forest classification algorithm trained on a cohort of at least 500 patient samples, the algorithm configured to receive normalized expression-level data and output a numerical prognostic score correlated with metastasis risk within 5 years; | Both (Literal disputed; DOE primary) | Helios's computational infrastructure includes one or more processors executing the ProstaSignal™ GBDT classification algorithm, trained on ~1,200 patient samples, which receives normalized expression data and outputs the ProstaScore (0–100) correlated with 5-year metastasis risk.
Same literal/DOE analysis as Claim 1, Element 1[b]: The 'random-forest classification algorithm' limitation is the central dispute. ProstaSignal™ uses gradient-boosted decision trees rather than random forests. DOE theory: function-way-result test is met; prosecution history estoppel analysis applies (see Claim 1 chart and Prosecution History Estoppel Analysis Memorandum). | Helios algorithm documentation (AEO); system architecture documents; same evidence as Claim 1[b] | Mirrors Claim 1[b] analysis. The 'processor' and 'system' framing does not change the core dispute over 'random-forest' vs. GBDT.
14[c] | and a reporting module configured to transmit the numerical prognostic score to a healthcare provider. | Literal | ProstaSignal™ generates a patient report containing the ProstaScore and risk classification (Low/Intermediate/High) and transmits it electronically (via secure portal and/or fax) to the ordering healthcare provider. Helios's reporting system constitutes a 'reporting module.' | Sample ProstaSignal™ patient reports; Helios reporting system documentation; provider portal screenshots | Literal infringement: The electronic reporting system directly meets this limitation.
[sheet: Claim 20]
Element No. | Claim Element / Limitation | Infringement Theory | ProstaSignal™ Corresponding Feature | Evidence / Source | Analysis / Notes
20[dependency] | The system of claim 14, wherein the reporting module is further configured to generate a report comprising the numerical prognostic score, a risk category classification, and a graphical representation of the patient's metastasis risk relative to a reference population. | Literal | The ProstaSignal™ patient report includes: (1) the numerical ProstaScore (0–100); (2) a risk category classification (Low, Intermediate, or High); and (3) a graphical chart showing the patient's ProstaScore plotted against a distribution curve of a reference patient population, visually illustrating the patient's relative metastasis risk.
All three sub-elements of this dependent claim are met by the ProstaSignal™ report format. | Sample ProstaSignal™ patient reports (multiple examples produced in discovery); Helios marketing materials showing sample report layout; ProstaSignal™ product brochure | Strong literal infringement. The ProstaSignal™ report format closely mirrors the claim language, including the graphical representation requirement.