Nigel DuffyDownload PDFPatent Trials and Appeals BoardFeb 9, 20222021002328 (P.T.A.B. Feb. 9, 2022) Copy Citation UNITED STATES PATENT AND TRADEMARK OFFICE UNITED STATES DEPARTMENT OF COMMERCE United States Patent and Trademark Office Address: COMMISSIONER FOR PATENTS P.O. Box 1450 Alexandria, Virginia 22313-1450 www.uspto.gov APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO. 13/412,887 03/06/2012 Nigel P. Duffy IH003CON 5269 54698 7590 02/09/2022 MOSER TABOADA 1030 BROAD STREET SUITE 203 SHREWSBURY, NJ 07702 EXAMINER SKIBINSKY, ANNA ART UNIT PAPER NUMBER 1631 NOTIFICATION DATE DELIVERY MODE 02/09/2022 ELECTRONIC Please find below and/or attached an Office communication concerning this application or proceeding. The time period for reply, if any, is set in the attached communication. Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the following e-mail address(es): docketing@mtiplaw.com llinardakis@mtiplaw.com PTOL-90A (Rev. 04/07) UNITED STATES PATENT AND TRADEMARK OFFICE ____________ BEFORE THE PATENT TRIAL AND APPEAL BOARD ____________ Ex parte NIGEL P. DUFFY Appeal 2021-002328 Application 13/412,887 Technology Center 1600 ____________ Before RICHARD M. LEBOVITZ, JEFFREY N. FREDMAN, and TIMOTHY G. MAJORS, Administrative Patent Judges. LEBOVITZ, Administrative Patent Judge. DECISION ON APPEAL The Examiner rejected 11-23 and 42 under 35 U.S.C. § 101 as lacking patent eligibility. Pursuant to 35 U.S.C. § 134(a), Appellant1 appeals from the Examiner’s decision to reject the claims. We have jurisdiction under 35 U.S.C. § 6(b). We AFFIRM. 1 We use the word “Appellant” to refer to “applicant” as defined in 37 C.F.R. § 1.42. Appellant identifies the real party in interest as Integral Health, Inc. Appeal Br. 3. Appeal 2021-002328 Application 13/412,887 2 STATEMENT OF THE CASE The Examiner rejected claims 11-23 and 42 in the Final Office Action under 35 U.S.C. § 101 because the claimed invention is directed to non- statutory subject matter. Final Act. 3. Claim 11, the only independent claim on appeal, is reproduced below (bracket numbering and indentations have been added for clarity and to reference the specific limitations in the claim): 11. A method for training a molecular properties model and predicting a property of interest for a test molecule, comprising, on a computer system: [1] obtaining a pseudo-partial ordering of molecules represented by a set of two or more ranked pairs of molecules, [1a] wherein the molecules within the pairs are ordered relative to one another based on the property of interest that is physically testable selected from the group consisting of reactivity, binding affinity, melting point, solubility, membrane permeability, toxicity, pKa, and combinations thereof, and [1b] wherein sources for measurements of the property comprise two or more separate protocols, [1c] wherein a subset of all pairs implied by one or more measurement sources is selected based on a probability that a difference in the property of interest falls outside experimental error, [1d] wherein the ranked pairs include one or more pairs of molecules ordered by measurements from such separate protocols; [2] generating a representation of the molecules included in the pseudo partial ordering of molecules that is appropriate for a selected machine learning algorithm; [3] providing as a training set the set of pairs of the pseudo partial ordering of molecules to the selected machine learning algorithm that optimizes a function of the pairs, [3a] wherein the molecules that comprise the pairs are ranked correctly in the obtained ranked pairs relative to the property of interest, and Appeal 2021-002328 Application 13/412,887 3 [3b] wherein, for at least a pair in the training set, said function penalizes an incorrectly ordered said pair implied by the function; [4] executing on the computer system the selected machine learning algorithm to generate a trained molecular properties model, [4a] wherein the measurements from two or more separate protocols contribute to the trained molecular properties model; [5] generating a representation of an additional molecule appropriate for the molecular properties model, which molecule is the test molecule; and [6] using the molecular properties model, generating on the computer system a prediction of a value of the property of interest for said test molecule. PRINCIPLES OF LAW Under 35 U.S.C. § 101, an invention is patent-eligible if it claims a “new and useful process, machine, manufacture, or composition of matter.” However, not every discovery is eligible for patent protection. Diamond v. Diehr, 450 U.S. 175, 185 (1981). “Excluded from such patent protection are laws of nature, natural phenomena, and abstract ideas.” Id. The Supreme Court articulated a two-step analysis to determine whether a claim falls within an excluded category of invention. Alice Corp. v. CLS Bank Int’l, 573 U.S. 208 (2014); Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66, 75-77 (2012). In the first step, it is determined “whether the claims at issue are directed to one of those patent-ineligible concepts.” Alice, 573 U.S. at 217. If it is determined that the claims are directed to an ineligible concept, then the second step of the two-part analysis is applied in which it is asked “[w]hat else is there in the claims before us?” Id. (alteration in original) (citation and quotation marks omitted). The Court explained that this step involves Appeal 2021-002328 Application 13/412,887 4 a search for an “‘inventive concept’”-i.e., an element or combination of elements that is “sufficient to ensure that the patent in practice amounts to significantly more than a patent upon the [ineligible concept] itself.” Alice, 573 U.S. at 217-18 (alteration in original) (citing Mayo, 566 U.S. at 75-77). Alice, relying on the analysis in Mayo of a claim directed to a law of nature, stated that in the second part of the analysis, “the elements of each claim both individually and ‘as an ordered combination’” must be considered “to determine whether the additional elements ‘transform the nature of the claim’ into a patent-eligible application.” Alice, 573 U.S. at 217 (citation omitted). The PTO published revised additional guidance on the application of 35 U.S.C. § 101. Eligibility Guidance, 84 Fed. Reg. 50. This guidance provides additional direction on how to implement the two-part analysis of Mayo and Alice. Step 2A, Prong One, of the Eligibility Guidance, looks at the specific limitations in the claim to determine whether the claim recites a judicial exception to patent eligibility. In Step 2A, Prong Two, the claims are examined to identify whether there are additional elements in the claims that integrate the exception in a practical application, namely, is there a “meaningful limit on the judicial exception, such that the claim is more than a drafting effort designed to monopolize the judicial exception.” 84 Fed. Reg. 54 (Prong Two). If the claim recites a judicial exception that is not integrated into a practical application, then as in the Mayo/Alice framework, Step 2B of the 2019 Eligibility Guidance instructs us to determine whether there is a Appeal 2021-002328 Application 13/412,887 5 claimed “inventive concept” to ensure that the claims define an invention that is significantly more than the ineligible concept itself. 84 Fed. Reg. 56. With these guiding principles in mind, we proceed to determine whether the claimed subject matter in this appeal is eligible for patent protection under 35 U.S.C. § 101. ARGUMENTS Appellant argues that the improvement to the technical field is using sources for measurements of the property of interest from [1b] “at least two or more separate protocols.” Appeal Br. 8-10. Appellant contends that using properties in the training module obtained from separate protocols was not routine, conventional, or well-understood. Id. at 9, 10 (“The Office has provided no evidence that such data from disparate sources was routinely or conventionally used in making molecular properties models. The claims are now amended to more exactingly require that data from separate data sources contributes to the model.”). Appellant repeats this same argument several times. Appeal Br. 10, 30, and 31. Appellant also contends that the Examiner did not meet the “Berkheimer burden”2 in establishing that it was 2 Berkheimer v. HP Inc., 881 F.3d 1360 (Fed. Cir. 2018) was decided by the Federal Circuit on February 8, 2018. On April 19, 2018, the PTO issued the Memorandum titled: “Changes in Examination Procedure Pertaining to Subject Matter Eligibility, Recent Subject Matter Eligibility Decision (Berkheimer v. HP, Inc.),” (“Berkheimer Memorandum”). Available at https://www.uspto.gov/sites/default/files/documents/memo-berkheimer- 20180419.PDF (last accessed Jan. 28, 2022). The Berkheimer Memorandum provided specific requirements for an Examiner to support with evidence that any additional claim element (or a combination of additional claim elements), beyond the abstract idea, represent well-understood, routine, or conventional activity. Appeal 2021-002328 Application 13/412,887 6 routine to use data from separate sources in the claimed process. See Appeal Br. 8, 11, 12, 30, and 31. Appellant also argues that the Examiner did not establish “that using such data as a pseudo-partial ordering of molecules represented by a set of two or more ranked pairs of molecules was routine or conventional.” Appeal Br. 11 (referencing step [1] of claim 11). See also Appeal Br. 31. To support these arguments, Appellant provided three declarations under 37 C.F.R. § 1.132 by Brandon Allgood, Ph.D. Dr. Allgood has a doctorate in physics. At the time, the declarations were executed, he was the Chief Technology Officer at Numerate, Inc. Dr. Allgood is not listed as an inventor of the claimed subject matter. The declarations are as follows: 1. Declaration under 37 C.F.R. § 1.132 by Brandon Allgood (executed July 22, 2015) (“Allgood Jul. 2015 Decl.”). 2. Declaration under 37 C.F.R. § 1.132 by Brandon Allgood (executed December 18, 2015) (“Allgood Dec. 2015 Decl.”). 3. Declaration under 37 C.F.R. § 1.132 by Brandon Allgood (executed November 12, 2018) (“Allgood Nov. 2018 Decl.”). We refer to the statements made in the declarations as the testimony of Dr. Allgood. ALLGOOD TESTIMONY Dr. Allgood, in three different declarations, testified about using measurements of a property of interest obtained from “two or more separate protocols” in the claimed method comprising “pseudo-partial ordering of molecules represented by a set of two or more ranked pairs of molecules.” Dr. Allgood explained: The method uses ranked pairs of molecules to train a machine learning algorithm operating on the computer. The reason using Appeal 2021-002328 Application 13/412,887 7 ranked pairs of molecules improves the computer function is that chemical and biological data for training sets is very noisy and very sensitive to the conditions under which the experiments are run. For example, that generated at Merck will not fully equate to that generated at Yale University. Generally, this means that using data from multiple sources is impossible. Even though the value of the measured quantity is sensitive to the assay conditions the relative measurements are very stable (e.g. the order). By training the models using ranked pairs, instead of absolute measurements the system becomes robust to this source of noise and allows one to ingest more data resulting in better models and better predictions. Allgood Jul. 2015 Decl. ¶ 4. Dr. Allgood further testified: With the method according to the invention, I have seen molecular properties models produced that are of good quality even where the data sources are diverse. Moreover, in these examples the data sources did not include enough data with sufficient experimental uniformity to allow a quality model to be built with traditional methods. If one were to use a [sic] such diverse set of data with prior art model building technique, the resulting model would be worse than if the model were trained only on a sub-set of the data … From my experience, in situations where uniform data sources are not abundant, the method of the current claims increases model quality. Allgood Dec. 2015 Decl. ¶ 8. Dr. Allgood also testified that he carried out experiments which showed that using “disparate data” from four different protocols and external data “created a better model” in comparison to using data from one protocol or just the four different protocols. Allgood Dec. 2015 Decl. ¶ 7. Dr. Allgood explained that a better model using the pseudo partial ordering was obtained because more data could be used in building the model. Id. ¶ 8. Appeal 2021-002328 Application 13/412,887 8 That is, using the data from different sources did not affect the quality of the model and permitted more data to be used. Dr. Allgood testified that he is “knowledgeable in the art of molecular modeling” and has “extensive contacts with users of this technology in the industry.” Allgood Nov. 2015 Decl. ¶ 9. He stated that if “someone else had been using pseudo-partial ordering of molecules in the time frame of June, 2004, l would have heard about it” and therefore it was not “routine, conventional or well understood” as of this date. Id. He made a similar statement about using data from disparate sources “via the claimed use of a pseudo-partial ordering of molecules. Id. ¶ 10. Dr. Allgood testified that the “multisource problem” of “utilizing data from disparate sources, i.e., different protocols or different laboratories” was known in the industry, but “it was not until the Applicant identified the claimed pseudo-partial ordering of molecules approach that it was solved.” Id. 10. 11. Dr. Allgood stated: Applicant has presented its methodology to pharmaceutical companies knowledgeable in molecular modeling, the presentations have elicited comments such as ‘once you read how Numerate has solved the multisource problem, it makes a lot of sense, because it is based on observations of biological assays and how they perform.’” Allgood Nov. 2015 Decl. ¶ 11. ANALYSIS We must determine whether the claim falls with a statutory category of invention. The claims are directed to a “method.” A method is also a “process,” one of the broad statutory categories of patent-eligible subject matter under 35 U.S.C. § 101. Because the claim falls into one of the statutory categories of patent-eligible subject matter, following the first step Appeal 2021-002328 Application 13/412,887 9 of the Mayo/Alice analysis, we proceed to Step 2A, Prong One, of the Eligibility Guidance. Step 2A, Prong One In Step 2A, Prong One, of the Eligibility Guidance, the specific limitations in the claim are examined to determine whether the claim recites a judicial exception to patent eligibility, namely, whether the claim recites an abstract idea, law of nature, or natural phenomenon. Each limitation in the claim must be searched to determine whether it recites a judicial exception. Eligibility Guidance, 84 Fed. Reg. at 54. Claim 11 has six steps, which we have numbered [1]-[6]. The claim is directed to a “method for training a molecular properties model and predicting a property of interest for a test molecule.” The claim provides a training set of ranked pairs of pseudo partial ordered molecules based on a property of interest (obtained in step [1] of the claim) to a machine learning algorithm (step [3]), and executes the algorithm (step [4]) to generate a trained molecular properties model. The model is used to predict a property of interest for a test molecule in steps [5] and [6] of the claim. The first step [1] of claim 11 comprises “obtaining a pseudo-partial ordering of molecules represented by a set of two or more ranked pairs of molecules.” The Specification discloses that a “partial order . . . is defined mathematically as a relation on a set with the properties of reflexivity i.e., antisymmetry i.e. (A ≤ B),(B ≤ A) ⇒ A= B and transitivity i.e., (A ≤ B),(B ≤ C) ⇒ (A ≤ C).” Spec. ¶ 35. The Specification explains that a “pseudo-partial order” or “PPO” is “defined herein as the relation on a set that can be viewed as a partial order for which antisymmetry does not hold and for which transitivity is does not hold.” The Specification further discloses that a PPO Appeal 2021-002328 Application 13/412,887 10 “includes a set of ranked pairs.” Id. ¶ 38. Because the Specification specifically defines PPO “mathematically,” we find that step [1] of claim 1 recites a “mathematical concept,” an abstract idea that is a judicial exception to patentability. Eligibility Guidance, 84 Fed. Reg. 52. The “wherein” clause [1a] of claim 11 recites that molecules are ordered based on a list of seven “physically testable” properties, which are the properties of interest that the model predicts for a test molecule. The claim does not require that the testing of the physical properties is done as part of the claimed method. The claim further recites that [1b] “wherein sources for measurements of the property comprise two or more separate protocols” and [1d] “wherein the ranked pairs include one or more pairs of molecules ordered by measurements from such separate protocols.” The measurement of the property is not a step in the claim. For example, the Specification describes using published data. Spec ¶ 57. There is no evidence in the record before us that the source from which the “physically testable” property measurement was obtained imparts a fingerprint or structure on the property. The Examiner further explained that the “source of the data does not change the nature of the data where data per se is abstract.” Final Act. 7. Limitation [1c] recites that, “wherein a subset of all pairs implied by one or more measurement sources is selected based on a probability that a difference in the property of interest falls outside experimental error.” Probability is mathematical concept. Consequently, we find that limitation [1c] also recites a judicial exception to patentability. Step [2] of the claim is “generating a representation of the molecules included in the pseudo partial ordering of molecules that is appropriate for a Appeal 2021-002328 Application 13/412,887 11 selected machine learning algorithm.” Appellant identifies paragraphs 13 and 41 as describing this limitation of the claim. Appeal Br. 5. Paragraph 13 merely recited substantially the same language in the claim without explaining how the “generating” is performed. Paragraph 41 discloses: Those skilled in the art will further recognize that PPOs may be represented using permutations of molecules, or sets of permutations of molecules. Further, when represented using sets of permutations of molecules, the permutations in the set may be assigned weights such that a weighted PPO is represented as a probability distribution over permutations of the molecules. The paragraph further describes other representations involving cross- products. Id. This step of the claim involves probability models (“a probability distribution over permutations of the molecules”), which is a mathematical concept and therefore an abstract idea. Citing paragraph 6 of the Specification, the Examiner also found that the recited representation of the molecules of step [2] also includes representing bonds between the atoms of a molecule, which the Examiner found includes representations “by mathematical equations for the energies of interactions between particles.” Final Act. 5 (citing the Allen pre-filing publication). The Examiner concludes that step [2] is therefore is a representation of information, which is abstract. Final Act. 5. A training set of the pseudo partial ordering of molecules is provided “to the selected machine learning algorithm that optimizes a function of the pairs” in step [3] of the claim. This information is used in step [4] to execute the machine learning algorithm “to generate a trained molecular properties model.” Limitation [4a] requires that “the measurements from two or more Appeal 2021-002328 Application 13/412,887 12 separate protocols contribute to the trained molecular properties model.” Limitation [4a] is just a restatement that the data in the subsequent steps from the two or more protocols is used. The Examiner found that the machine learning algorithm recited in step [4] of the claim is a mathematical concept. Final Act. 3-4. The Examiner explained that dependent claim 15 identifies the “selected machine algorithm” of claim 11 as being selected from a list including a variant of a RankBoost algorithm. Id. at 3. The Examiner cited a pre-filing date publication as evidence that the RankBoost algorithm is a series of mathematical steps and therefore a mathematical concept. Id. at 3-4. After the trained molecular properties model is generated, a prediction of a value of the property of interest used to train the model is generated from a test molecule is accomplished in steps [5] and [6] of the claim. The Examiner found this step recites an abstract idea because it produces “information about molecules.” Final Act. 5. To sum it up, at least limitations [1], [1c], [2], and [4] of claim 11 recite abstract ideas. Appellant argues: [I]t is only the well-known steps of molecular properties modeling that are stated somewhat abstractly. That which makes the method novel and unobvious is stated concretely. It is not necessary that a claim recite each and every element needed for the practical realization of the claimed subject matter. Appeal Br. 23. First, this argument appears to be an admission that the claim, as the Examiner explained and discussed above, recites one or more abstract idea. Second, whether the claim is “novel” or “unobvious” is not at issue with this Appeal 2021-002328 Application 13/412,887 13 rejection. The rejection is under § 101, not § 102 or § 103. The rejection is based on the Examiner’s finding that the claim recites an abstract idea. A claim can recite new and unobvious subject matter and still be directed to an abstract idea. Dr. Allgood in his declaration also admits that the steps of the claim involve mathematical concepts. Dr. Allgood, in summarizing certain steps of the claim, testified that the steps “tie the mathematical operation (visualizing, predicting properties of interest) to the processor’s ability to better, more robustly process many sources of data to generate the prediction.” Allgood Jul. 2015 Decl. ¶ 6. Because we conclude that claim 1 recites a judicial exception to patent eligibility, we proceed to Step 2A, Prong Two of the analysis. Step 2A, Prong Two Prong Two of Step 2A of the Eligibility Guidance asks whether there are additional elements that integrate the exception into a practical application. As explained in the Eligibility Guidance, integration may be found when an additional element “reflects an improvement in the functioning of a computer, or an improvement to other technology or technical field” or “applies or uses the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception.” Eligibility Guidance, 84 Fed. Reg. at 55. Appeal 2021-002328 Application 13/412,887 14 Pseudo-partial ordering of molecules Step [1] of claim 11 recites “obtaining a pseudo-partial ordering of molecules represented by a set of two or more ranked pairs of molecules,” where the molecules are ranked relative to each other by a property of interest (limitation [1a]) and where the sources for measurements of the property of interest comprise two or more separate protocols (limitation [1b] and [1d]). These measurements, from at least two different protocols, are used in training the machine algorithm (limitation [4a]). We must first address whether using property measurements from two or more separate protocols in training a machine learning algorithm to generate a molecular model is a patentable improvement to a technical field. The Specification explains that measurements made from different protocols may result in inconsistent values for a property of the same molecule may be reported. Spec. ¶¶ 7, 46. “For example, measurements obtained from one lab or using one experimental protocol may consistently assign higher values for a property of interest to a particular molecule than others.” Id. ¶ 7. We understand the “separate protocols” to at least include measurements using different protocols (such as different conditions) or measurements using the same protocol but in different laboratories. The Specification discloses that applying the claimed pseudo-partial ordering of ranked molecules makes the data “fairly consistent, regardless of source.” Id. ¶ 47. Based on Dr. Allgood’s testimony, we understand the asserted improvement to be in the technical field of molecular model building. The improvement, as described in the Allgood declarations, is the use of pseudo- partial ordered molecules to train a machine learning algorithm to predict a property of interest of a test molecule. Allgood Jul. 2015 Decl. ¶ 4 Appeal 2021-002328 Application 13/412,887 15 (reproduced above). According to Dr. Allgood, this improvement permits property measurements from disparate sources (“two or more separate protocols”) to be utilized in the training set without affecting the quality of the model. Allgood Jul. 2015 Decl. ¶ 4 (reproduced above); Allgood Dec. 2015 Decl. ¶ 8 (reproduced above). As we understand it, the improvement is important because it permits more data (namely data from different sources and multiple protocols) to be used in training the machine learning algorithm, and thus results in a better quality model for predicting a property of interest. While the improvement to molecular model building may be significant, not all advances are patent eligible. As held in Association for Molecular Pathology v. Myriad Genetics Inc., 569 U.S. 576, 591 (2013), “[g]roundbreaking, innovative, or even brilliant discovery does not by itself satisfy the § 101 inquiry.” In Electronic Communication Technologies, LCC v. ShopperChoice.com, 958 F.3d 1178, 1181-1182 (Fed. Cir. 2020), an advance in security and authentication was found in ineligible for a patent because the improvement was in the “realm of abstract ideas.” The court in SAP Am., Inc. v. InvestPic, LLC, 898 F.3d 1161, 1170 (Fed. Cir. 2018) explained that “[u]nder the principles developed in interpreting § 101, patent law does not protect such claims, without more, no matter how groundbreaking the advance.” The improvement asserted by Appellant is the application of pseudo- partial ordering of molecules to a machine learning algorithm. Allgood Nov. 2015 Decl. ¶¶ 9, 10 (cited supra at 7). However, as explained under Step 2A, Prong One, the “pseudo-partial ordering of molecules represented by a set of two or more ranked pairs of molecules” is a mathematical concept, one of the groupings of subject matter that courts have found to be in the realm of Appeal 2021-002328 Application 13/412,887 16 abstract ideas. Eligibility Guidance, 84 Fed. Reg. 52. The machine learning algorithm also involves a mathematical concept and abstract idea. See supra at 12. To the extent that these steps provide the asserted improvement, they are part of the abstract idea and “[t]he abstract idea itself cannot supply the inventive concept.” Trading Tech. Int’l, Inc. v. IBG LLC, 921 F.3d 1378, 1385 (Fed. Cir. 2019). The steps do not integrate the abstract idea into a practical application because the steps represent the abstract idea, itself. Utilizing property measurements from different sources, as recited in limitations [1b], [1d], and [4a], does not alter the abstract nature of the claimed process of using the set of pairs of the pseudo partial ordering of molecules as a training set for the selected machine learning algorithm. The claim does not recite a step in which the measurements are collected. Instead, the measurements are simply used as the data to accomplish the pseudo-partial ordering of the molecules. The data has no memory or fingerprint from where it was obtained; the property measurement is simply the information that is used to generate the ranked pairs for training the machine learning algorithm. See Final Act. 7. In essence, the claims are like those in SAP America, 898 F.3d at 1167, where the court found that claims involving the statistical analysis of investment data were focused on “selecting certain information, analyzing it using mathematical techniques, and reporting or displaying the results of the analysis” which they determined to be “all abstract.” Here, information about a property of interest of molecules (limitation [1b]) is analyzed by employing pseudo-partial ordering (step [1]) and a machine learning algorithm (step [3]), both of which are mathematical techniques, to produce a molecular properties model (step [4]) that predicts information about the Appeal 2021-002328 Application 13/412,887 17 property of interest of a test molecule (step [6]). Thus, the rejected claims analyze information using mathematical techniques to obtain additional information about the property of a molecule. The use of existing information in a mathematical algorithm was found patent ineligible in Digitech Image Technologies, LLC v. Electronics for Imaging, Inc., 758 F.3d 1344, 1348 (2014). The claims in Digitech were directed to “a process of taking two data sets and combining them into a single data set, the device profile. The two data sets are generated by taking existing information-i.e., measured chromatic stimuli, spatial stimuli, and device response characteristic functions-and organizing this information into a new form.” Digitech, 758 F.3d at 1351. The court found that the method claims were drawn to “an abstract idea because it describes a process of organizing information through mathematical correlations and is not tied to a specific structure or machine.” Id. at 1350. Furthermore, the court held: The above claim thus recites an ineligible abstract process of gathering and combining data that does not require input from a physical device. As discussed above, the two data sets and the resulting device profile are ineligible subject matter. Without additional limitations, a process that employs mathematical algorithms to manipulate existing information to generate additional information is not patent eligible. Id. at 1351. The claims in this case are analogous to those in Digitech because there is no input of data from a physical device, but rather the claim involves the manipulation of existing data obtained from separate protocols to generate additional data employing mathematical concepts. Appeal 2021-002328 Application 13/412,887 18 Dr. Allgood’s testimony about creating a “better model” is based on the discovery that the pseudo partial ordering of step [1] of claim 11 enables data sets to be utilized from different protocols. See supra at 7 (citing Allgood Nov. 2015 Decl. ¶¶ 10, 11). The solution described by Dr. Allgood for the multi-source problem is the discovery that a mathematical concept, pseudo partial ordering of molecules, improves machine learning, another mathematical concept. See supra at 7-8 (discussing Allgood Nov. 2015 Decl. ¶¶ 9-11). None of this is sufficiently outside the realm of abstract ideas to impart eligibility to the claims on appeal. Improvement to computer Dr. Allgood testified that, “using ranked pairs of molecules improves the computer function.” Allgood Jul. 2015 Decl. ¶ 4 (see supra at 6). Dr. Allgood stated that the “reason” computer function is improved is “that chemical and biological data for training sets is very noisy and very sensitive to the conditions under which the experiments are run.” Id. Dr. Allgood explained: “By training the models using ranked pairs, instead of absolute measurements the system becomes robust to this source of noise and allows one to ingest more data resulting in better models and better predictions.” Id. Citing claim 11 before it was amended to recite [1b] “wherein sources for measurements of the property comprise two or more separate protocols,” Dr. Allgood states the steps “tie the mathematical operation (visualizing, predicting properties of interest) to the processor’s ability to better, more robustly process many sources of data to generate the prediction,” which “thus improve the functioning of the claimed computer itself.” Allgood Jul. 2015 Decl. ¶¶ 5, 6. Appellant makes the same argument, stating: Appeal 2021-002328 Application 13/412,887 19 By the evidence of record in this case, Appellant improves computer functionality by making molecular properties modeling robust even when the needed data comes from disparate sources, utilizing different experimental protocols. Appellant does this by a “specific technique that departs from earlier approaches,” namely using pairs of a pseudo partial ordering of molecules, and using data from different protocols. The record has no evidence of these claim recitals being used in training a molecular properties model prior to Appellant's filing. Appeal Br. 19. In sum, Appellant reasons that “computer function” is improved by making “the processor’s ability to better, more robustly process many sources of data to generate the prediction” and “by making molecular properties modeling robust even when the needed data comes from disparate sources.” We are not persuaded by Appellant’s argument that using pseudo- partial ordering in a machine learning algorithm to make it more “robust” in predicting property information about a molecule is an improvement to “computer function.” Dr. Allgood explains that the improvement in terms of the “ingest[ion]” of “more data resulting in better models and better predictions” by using “ranked pairs of molecules to train a machine learning algorithm operating on the computer.” Allgood Jul. 2015 Decl. ¶ 4 (see supra at 6). Dr. Allgood ascribes this improvement to the computer processor, but the processor is simply executing the machine learning algorithm. Dr. Allgood did not provide evidence that processor used to process the data is anything more than a generic processor.3 3 Claim 11 recites a “computer system.” For the purpose of the analysis, we equate the claimed “computer system” with the “processor” discussed by Dr. Allgood. Appeal 2021-002328 Application 13/412,887 20 Appellant cites to Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), among other cases, to support their argument that computer functionality is improved. Appeal Br. 7. In Enfish, 1339 (Fed. Cir. 2016), the claims were found to recite patent eligible subject matter because “the self- referential table recited in the claims on appeal is a specific type of data structure designed to improve the way a computer stores and retrieves data in memory.” See also Visual Memory LLC v. NVIDIA Corp., 867 F.3d 1253, 1259 (Fed. Cir. 2017) (“Our review of the ’740 patent claims demonstrates that they are directed to an improved computer memory system, not to the abstract idea of categorical data storage.”). Appellant has not explained in this case how “the mathematical operation (visualizing, predicting properties of interest)” improves the function of the computer processor in the same way in Enfish and Visual Memory. Allgood Jul. 2015 Decl. ¶¶ 5, 6. To the contrary, claim 11 lacks eligibility for reasons similar to those articulated in In re Board of Trustees of Leland Stanford Junior University, 989 F.3d 1367, 1373 (Fed. Cir. 2021): Even accepting the argument that the claimed process results in improved data, we are not persuaded that claim 1 is not directed to an abstract mathematical calculation. Synopsys, Inc. v. Mentor Graphics Corp., 839 F.3d 1138, 1151 (Fed. Cir. 2016) (“[A] claim for a new abstract idea is still an abstract idea.”) Appellant also argues that, “there is no general rule that including arguably broadly and generically recited mathematical steps precludes eligibility.” Appeal Br. 27. We agree. But when the asserted improvement is a mathematical step, the improvement is to the abstract idea, not an additional element in the claim, beyond the abstract idea, and therefore the claim is ineligible for a patent under § 101. Appeal 2021-002328 Application 13/412,887 21 Because we find that the abstract idea is not integrated into a practical application under Step 2, Prong Two, we proceed to Step 2B of the analysis Step 2B Step 2B of the Eligibility Guidance asks whether there is an inventive concept. In making this Step 2B determination, we must consider whether there are specific limitations or elements recited in the claim “that are not well-understood, routine, conventional activity in the field, which is indicative that an inventive concept may be present,” or whether the claim “simply appends well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception, which is indicative that an inventive concept may not be present.” Eligibility Guidance, 84 Fed. Reg. 56 (footnote omitted). Dr. Allgood testified that it was not “routine, conventional or well understood” as June 2004 to use disparate data from different protocols in the claimed pseudo-partial ordering of molecules to create a molecular properties model. See supra 7-8 (Allgood Nov. 2018 Decl. ¶ 9-11; Allgood Jul. 2015 Decl. ¶ 4). Still, the crux of the improvement is mathematical and abstract, and therefore, whether that alleged improvement is novel and unobvious, it is insufficient as the basis of an inventive concept to establish patent eligibility of the claims. As explained in the 2019 Eligibility Guidance, 84 Fed. Reg. 55, 56, inventive concept under Step 2B must be provided by an “additional element” in the claim that is “beyond the judicial exception.” Dr. Allgood did not identify an additional element that satisfies the inventive concept prong of the eligibility analysis. Because the claim elements described by Dr. Allgood and Appellant are abstract and not additional elements of the claim, the Examiner did not Appeal 2021-002328 Application 13/412,887 22 have the burden under these circumstances to demonstrate that such claim elements are well-understood, routine, and conventional. Appellant’s attempt to place this burden on the Examiner conflates patent eligibility under § 101 with patentability under §§ 102 and 103. Indeed, [t]he ‘novelty’ of any element or steps in a process, or even of the process itself, is of no relevance in determining whether the subject matter of a claim falls within the § 101 categories of possibly patentable subject matter.” Diamond v. Diehr, 450 U.S. 175, 188-89, 101 S.Ct. 1048, 67 L.Ed.2d 155 (1981) (emphasis added); see also Mayo, 132 S.Ct. at 1303-04 (rejecting “the Government’s invitation to substitute §§ 102, 103, and 112 inquiries for the better established inquiry under § 101”). Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1315 (Fed. Cir. 2016). Appellant did not provide evidence the claimed “computer system” is anything more than a generic computer that implements the pseudo-ordering of molecules, machine learning algorithm, etc., in the conventional and well- known way that computers ordinary function. To qualify as “a patent- eligible improvement,” the invention must be directed to a specific improvement in the computer’s functionality, not simply to use of the computer “as a tool” to implement an abstract idea. Customedia Techs., LLC v. Dish Network Corp., 951 F.3d 1359, 1363-1364 (Fed. Cir. 2020). Here, the invention falls into the latter category. It focuses on using a general purpose computer to carry out the abstract idea. Consequently, we do not discern an inventive concept in how the computer system operates. For the foregoing reasons, the rejection of claim 11 is affirmed. Claims 12-23 and 42 fall with claim 11 because separate reasons for their patent ability were not provided. 37 C.F.R. § 41.37(c)(1)(iv). Appeal 2021-002328 Application 13/412,887 23 DECISION In summary: Claims Rejected 35 U.S.C. § Reference(s)/Basis Affirmed Reversed 11-23, 42 101 Patent Eligibility 11-23, 42 TIME PERIOD No time period for taking any subsequent action in connection with this appeal may be extended under 37 C.F.R. § 1.136(a)(1)(iv). AFFIRMED Copy with citationCopy as parenthetical citation