UCB BIOPHARMA SPRL

Patent Trials and Appeals Board

Nov 16, 2021

2021002263 (P.T.A.B. Nov. 16, 2021)

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.
15/415,758 01/25/2017 Kunal MALHOTRA 505.1010 4103
23280 7590 11/16/2021
Davidson, Davidson & Kappel, LLC
589 8th Avenue
22nd Floor
New York, NY 10018
EXAMINER
LEE, ANDREW ELDRIDGE
ART UNIT PAPER NUMBER
3626
NOTIFICATION DATE DELIVERY MODE
11/16/2021 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):
ddk@ddkpatent.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________

Ex parte KUNAL MALHOTRA, SUNGTAE AN, JIMENG SUN,
MYUNG CHOI, CYNTHIA DILLEY, CHRIS CLARK,
JOSEPH ROBERTSON, and EDWARD HAN-BURGESS
____________

Appeal 2021-002263
Application 15/415,758
Technology Center 3600
____________

Before NINA L. MEDLOCK, BRUCE T. WIEDER, and
BRADLEY B. BAYAT, Administrative Patent Judges.

MEDLOCK, Administrative Patent Judge.

DECISION ON APPEAL

Appeal 2021-002263
Application 15/415,758
2
STATEMENT OF THE CASE
Appellant1 appeals under 35 U.S.C. § 134(a) from the Examiner’s
final rejection of claims 1–3 and 5–26. We have jurisdiction under
35 U.S.C. § 6(b).
We REVERSE.
CLAIMED INVENTION
The Specification states, “[t]he present disclosure relates generally to
a method of predicting optimal treatment regimens and more specifically to
a method of predicting optimal treatment regimens for epilepsy patients”
(Spec. ¶ 1).
Claims 1, 15, 21, and 26 are the independent claims on appeal.
Claim 1, reproduced below with bracketed notations added, is illustrative of
the claimed subject matter:
1. A method of training a machine learning algorithm
for predicting the efficacy of anti-epilepsy drug treatment
regimens comprising:
[(a)] providing electronic health records data;
[(b)] constructing a patient cohort from the electronic
health records data by selecting patients based on a defined target
variable indicating anti-epilepsy drug treatment regimen
efficacy;
[(c)] constructing a set of features found in or derived
from the electronic health records data;

1 We use the term “Appellant” to refer to “applicant” as defined in 37 C.F.R.
§ 1.42. Our decision references Appellant’s Appeal Brief (“Appeal Br.,”
filed September 22, 2020) and Reply Brief (“Reply Br., filed February 9,
2021), and the Examiner’s Answer (“Ans.,” mailed December 28, 2020) and
Final Office Action (“Final Act.,” mailed April 6, 2020). Appellant
identifies UCB BIOPHARMA SRL as the real party in interest (Appeal
Br. 2).
Appeal 2021-002263
Application 15/415,758
3
[(d)] selecting features by electronically processing the
set of constructed features based on the patient cohort to select a
subset of the constructed features that are predictive for anti-
epilepsy drug treatment regimen efficacy for inclusion in
predictive computerized models configured for generating
predictions representative of efficacy for a plurality of anti-
epilepsy drug treatment regimens, the selecting of features
including selecting respective unique features for each of the
plurality of anti-epilepsy drug treatment regimens and selecting
different numbers of features for at least two of the plurality of
anti-epilepsy drug treatment regimens; and
[(e)] training each of the predictive computerized models
to generate predictions representative of efficacy for a plurality
of anti-epilepsy drug treatment regimens for the patients based
on the defined target variable indicating anti-epilepsy drug
treatment regimen efficacy, each of the predictive computerized
models being specific to a respective different one of the anti-
epilepsy drug treatment regimens and configured for generating
predictions representative of the efficacy of the respective
different one of the anti-epilepsy drug treatment regimens for a
patient based on the selected respective unique features.
REJECTIONS
Claims 1–3, 8, and 14 are rejected under 35 U.S.C. § 103 as
unpatentable over De Bruin et al. (US 2011/0119212 A1, published May 19,
2011) (“De Bruin”), Kheifetz et al. (US 2017/0140109 A1, published
May 18, 2017) (“Kheifetz”), and Dilorenzo et al. (US 2007/0150025 A1,
published June 28, 2007) (“Dilorenzo”).
Claims 5, 9, 12, and 13 are rejected under 35 U.S.C. § 103 as
unpatentable over De Bruin, Kheifetz, Dilorenzo, and Ebadollahi et al.
(US 2015/0019239 A1, published Jan. 15, 2015 (“Ebadollahi”).
Claims 6 and 7 are rejected under 35 U.S.C. § 103 as unpatentable
De Bruin, Kheifetz, Dilorenzo, and Pestian et al (US 2016/0180041 A1,
published June 23, 2016) (“Pestian”).
Appeal 2021-002263
Application 15/415,758
4
Claims 10 and 11 are rejected under 35 U.S.C. § 103 as unpatentable
over De Bruin, Kheifetz, Dilorenzo, and Benner et al. (US 10,108,975 B1,
issued Oct. 23, 2018) (“Benner”).
Claims 15–17, 19, and 20 are rejected under 35 U.S.C. § 103 as
unpatentable over De Bruin, Kheifetz, Dilorenzo, and Narain et al.
(US 2016/0171383 A1, published June 16, 2016) (“Narain”).
Claim 18 is rejected under 35 U.S.C. § 103 as unpatentable over
De Bruin, Kheifetz, Dilorenzo, Narain, and Miller et al.
(US 2016/0301691 A1, published Oct. 13, 2016) (“Miller”).
Claims 21, 22, 24, and 25 are rejected under 35 U.S.C. § 103 as
unpatentable over De Bruin, Kheifetz, Dilorenzo, Narain, and Moon et al.
(US 2018/0046918 A1, published Feb. 15, 2018) (“Moon”).
Claim 23 is rejected under 35 U.S.C. § 103 as unpatentable over
De Bruin, Kheifetz, Dilorenzo, Narain, Moon, and Miller.
Claim 26 is rejected under 35 U.S.C. § 103 as unpatentable over
De Bruin, Kheifetz, Dilorenzo, Narain, and Pestian.
ANALYSIS
Independent Claim 1 and Dependent Claims 2, 3, 8, and 14
In rejecting claim 1 under 35 U.S.C. § 103, the Examiner cited
De Bruin as disclosing substantially all of the limitations of claim 1 (Final
Act. 7–10). But the Examiner acknowledged that De Bruin does not
explicitly teach, inter alia, “the selecting of features including selecting
respective unique features for each of the plurality of anti-epilepsy drug
treatment regimens and selecting different numbers of features for at least
two of the plurality of anti-epilepsy drug treatment regimens,” i.e., step (d)
Appeal 2021-002263
Application 15/415,758
5
of claim 1 (id. at 10); and the Examiner cited Kheifetz to cure this deficiency
(id. at 10–12 (citing Kheifetz ¶¶ 47, 57, 59–62, 68, 79–82, Figs. 4, 5)).
Responding to Appellant’s arguments, and outlining, in the Answer,
the bases for the rejection, the Examiner cites paragraph 47 of Kheifetz as
teaching the use of disease progression models (DPMj), each of which
corresponds to a particular treatment protocol (TPk) (Ans. 4). The Examiner
notes that claim 1 does not “state what the features are, just that they are
unique for each treatment regimen and contain a different number of features
for at least [two] of the treatment regimens” (id. at 4–5). And the Examiner
concludes that, under a broadest reasonable interpretation, this limitation is
taught by “the use of Pharmacokinetics (PK) and Pharmacodynamics (PD)
as the features for the disease progression models, which are specific for
each of the drugs used in the treatment regimens which are each specific to
each disease progression model” (id. at 5 (citing Kheifetz ¶¶ 59–62, Figs. 4
and 5); see also Final Act. 30 (“The Examiner notes each model (DPMj) is
specific to a particular treatment protocol, the features are the
Pharmacokinetics (PK) and Pharmacodynamics (PD) of the drugs in the
treatment protocols, these are unique features to each treatment protocol.”)).2
The Examiner asserts that “Kheifetz teaches that each disease
progression model has features for one specific treatment protocol” and that
“models can have a different [number] of drugs” (Ans. 5.). Thus,
referencing Figures 4 and 5, which the Examiner characterizes as showing

2 Kheifetz discloses “[t]he provision of each of the reference/basic disease
progression models DPMj (stage 100) may include adjusting a population
model for a specific treatment protocol” and that the population model
includes “Pharmacokinetics (PK) and Pharmacodynamics (PD) of relevant
drugs 112 which is the data about drugs typically used in the treatment of the
specific medical condition” (Kheifetz ¶ 59).
Appeal 2021-002263
Application 15/415,758
6
one model (for determining the efficacy of two drugs) having features PDCIS,
PDPEM, CCIS, and CPEM, unique to the model, the Examiner reasons that
because “the models for the drugs used alone would not contain the of PD or
C for the other drug respectively, and therefore would be a model with 2 less
features” (id.), “[t]his reads on at least 2 models with [a] different number of
features which are also unique to each . . . regimen, which teaches what is
required” (id.).
The difficulty with the Examiner’s analysis, as Appellant observes, is
that the Examiner is ostensibly interpreting “features,” as called for in
claim 1, not as measureable properties, but as the measurements of those
properties (Reply Br. 3 (explaining that in machine learning “a feature is an
individual measurable property or characteristic of a phenomenon being
observed”; for example, age is a feature but age 39 is not a different feature
from age 40; instead both are different measurements of the same feature)).
We agree with Appellant that two different models, including different PK
and PD values, may include different measurements of those features; but
the two models do not include different features, as called for in claim 1.
Instead, they include the same features (id. at 3–4).3
In view of the foregoing, we do not sustain the Examiner’s rejection
of claim 1 under 35 U.S.C. § 103. For the same reasons, we also do not
sustain the Examiner’s rejection of dependent claims 2, 3, 8, and 14. Cf. In
re Fritch, 972 F.2d 1260, 1266 (Fed. Cir. 1992) (“dependent claims are

3 The Examiner notes that, in Kheifetz, models can have a different number
of drugs (Ans. 5). But, as best understood, the Examiner does not draw any
correlation between the number of drugs and feature selection.
Appeal 2021-002263
Application 15/415,758
7
nonobvious if the independent claims from which they depend are
nonobvious”).
Dependent Claims 5–7 and 9–13
Each of claims 5–7 and 9–13 depends, directly or indirectly, from
independent claim 1. The rejections of these dependent claims do not cure
the deficiencies in the Examiner’s rejection of independent claim 1.
Therefore, we do not sustain the Examiner’s rejections under 35 U.S.C.
§ 103 of dependent claims 5–7 and 9–13 for the same reasons set forth
above with respect to claim 1.
Independent Claim 15 and Dependent Claims 16, 17, 19, and 20
Independent claim 15 includes language substantially similar to the
language of independent claim 1, and stands rejected based on the same
rationale applied with respect to claim 1 (see Final Act. 28–32). Therefore,
we do not sustain the Examiner’s rejection under 35 U.S.C. § 103 of
independent claim 15, and claims 16, 17, 19, and 20, which depend
therefrom, for the same reasons set forth above with respect to claim 1.
Dependent Claim 18
Claim 18 depends from independent claim 15. The rejection of this
dependent claim does not cure the deficiencies in the Examiner’s rejection of
independent claim 15. Therefore, we do not sustain the Examiner’s rejection
under 35 U.S.C. § 103 of dependent claim 18 for the same reasons set forth
above with respect to claim 15.
Appeal 2021-002263
Application 15/415,758
8
Independent Claim 21 and Dependent Claims 22, 24, and 25
Independent claim 21 recites in part:
21. A computerized method for generating anti-
epilepsy drug treatment regimen efficacy predictions
comprising:
providing a pre-trained machine learning algorithm for
predicting efficacy of anti-epilepsy drug treatment regimens,
the pre-trained machine learning algorithm including pre-
trained anti-epilepsy drug treatment regimen efficacy prediction
models[;]
* * * *
requesting, via a client, formatted electronic medical
records data for a patient from an electronic medical records
database;
implementing a query to translate medical codes in the
formatted electronic medical records data into a coding system
used to communicate with the pre-trained anti-epilepsy drug
treatment regimen efficacy prediction models; [and]
mapping features from the formatted electronic medical
records data into a further format, the mapping including
converting information for both anti-epileptic drugs and non-
anti-epileptic drugs prescribed to the patient into a feature
matrix as event data identified by a prefix . . . .
Appeal Br., Appendix A 5–6.
In rejecting claim 21 under 35 U.S.C. § 103, the Examiner cited
Dilorenzo as disclosing “mapping features from the formatted electronic
medical records data into a further format, the mapping including converting
information for both antiepileptic drugs and non-anti-epileptic drugs
prescribed to the patient into a feature matrix,” as recited in claim 21 (Final
Act. 43 (citing Dilorenzo ¶¶ 28, 29, 62–69, 87–89, 92–96, 111, 112, 118,
128, 129, and 142)). But, the Examiner acknowledged that Dilorenzo does
not disclose that the information is converted into “event data identified by a
prefix” (id. at 46). The Examiner cited Moon to cure this deficiency (id.
Appeal 2021-002263
Application 15/415,758
9
(citing Moon ¶¶ 3, 5, 6, 20, 31–33, 44; Figs. 1, 2)). And the Examiner
concluded that
[o]ne of ordinary skill in the art before the effective filing date
would have found it obvious to include using a prefix to identify
event data as taught by Moon within the method of using a pre-
trained anti-epilepsy treatment regimen efficacy model on
received EMR data which maps drugs prescribed to a patient to
a feature matrix as taught by De Bruin, Kheifetz, and Dilorenzo
with the motivation of “improve the predictions of the model for
particular data items” (Moon: paragraph [0003]).
Id. at 47.
Moon4 is titled “Aggregate Features for Machine Learning,” and
discloses that “[i]mplementations provide a flexible infrastructure for . . .
using historical data to generate input features for a machine learned model,”
which can be used “to improve the predictions of the model for particular
data items (e.g., for personalized recommendations)” (Moon ¶ 3). Moon
describes that “aggregate features” are “data records that store a value for an
aggregate function (count, average, etc.) of a feature over some window of
time,” and that these aggregate features can be used to “provide more
relevant content for a particular user with low latency, e.g., to find real-time
information and to provide query suggestions (e.g., type-ahead, spelling,
related searches) that are also fresh and real-time” (id.).
Referencing Figure 2 in paragraph 20, Moon describes an aggregate
feature definition scheme, where a prefix 205 is used to identify aggregate
feature entries for an aggregate group (id. ¶ 20). And Moon explains that if
prefix 205 is “‘user_aggregate’ a query . . . engine may locate aggregate

4 On its face, Moon shows Twitter, Inc. as the applicant.
Appeal 2021-002263
Application 15/415,758
10
feature entries for this group by identifying records that match the glob
pattern ‘user_aggregate*” (id.).
Appellant does not challenge the Examiner’s findings with regard to
Dilorenzo; instead Appellant argues that Moon’s “prefix 205 allows real-
time records to be retrieved in response to typing by a user to provide the
user with suggestions for a user entering a query into Twitter” and “would
not have provided one of ordinary skill in the art at the time of the present
invention with any reason to have modified the method of De Bruin,” which
“is not at all concerned with real-time suggestions in response to a query”
(Appeal Br. 17–18). Appellant, thus, charges that the Examiner is “clearly
selectively combining these references based on improper hindsight bias”
and, moreover, that the Examiner’s stated motivation for modifying
De Bruin in view of Moon (i.e., to “improve the predictions of the model for
particular data items”) “completely lacks a rational[ ] underpinning,” i.e.,
“[t]here is absolutely no teaching in Moon that prefixes 205 improves the
prediction” (id. at 18–19).
Responding to Appellant’s argument, the Examiner asserts that “the
breath [sic] of the claim only requires that data [are] mapped to an event
identifiable with a prefix” and that “Moon teaches[,] at paragraphs [0020]
and [0031]–[0033], a prefix element identifying features which are in a
feature matrix” (Ans. 6). The Examiner concludes, “[t]herefore, . . . the
combination of De Bruin, Dilorenzo and Moon teaches what is required
under the broadest reasonable interpretation” (id. at 6–7). But, as Appellant
observes, the Examiner does not adequately explain why a person of
ordinary skill in the art would have been prompted to “appl[y] the prefixes
used in Moon’s algorithm for generating ‘personalized recommendations’
for Twitter queries to the . . . ‘mapped features’ of the . . . De Bruin-
Appeal 2021-002263
Application 15/415,758
11
Dilorenzo-Narain combination that are run through the anti-epilepsy drug
treatment regimen efficacy prediction models” (Reply Br. 6).
The Examiner has not established, on the present record, that a person
of ordinary skill in the art at the time of Appellant’s invention would have
had an apparent reason to combine De Bruin, Dilorenzo, and Moon, as the
Examiner proposes, to arrive at the claimed invention. See In re Kahn,
441 F.3d 977, 988 (Fed. Cir. 2006) (“[R]ejections on obviousness grounds
[require] some articulated reasoning with some rational underpinning to
support the legal conclusion of obviousness”) (cited with approval in KSR
Int’l Co. v. Teleflex Inc., 550 U.S. 398, 418 (2007). Therefore, we do not
sustain the Examiner’s rejection of independent claim 21 under 35 U.S.C.
§ 103. For the same reasons, we also do not sustain the rejection of
dependent claims 22, 24, and 25.
Dependent Claim 23
Claim 23 depends from independent claim 21. The rejection of this
dependent claim does not cure the deficiency in the Examiner’s rejection of
independent claim 21. Therefore, we do not sustain the Examiner’s rejection
under 35 U.S.C. § 103 of dependent claim 23 for the same reasons set forth
above with respect to claim 21.
Independent Claim 26
Independent claim 26 is directed to a computer platform for
generating drug treatment regimen efficacy predictions, and recites that the
computer platform comprises, inter alia,
a client configured for interfacing with a data interface
server, the data interface server configured to request formatted
electronic medical records data for a patient from an electronic
medical records database; [and]
Appeal 2021-002263
Application 15/415,758
12
a feature mapping tool configured for mapping features
from the formatted electronic medical records data into a further
format, the feature mapping tool configured for aggregating all
raw epilepsy diagnosis codes and all raw convulsion codes in the
formatted electronic medical records data of the patient into a
higher level code in a feature matrix using a medical
classification scheme . . . .
Appeal Br., Appendix A 7.
In rejecting independent claim 26 under 35 U.S.C. § 103, the
Examiner relied on Pestian as disclosing a feature mapping tool, as called for
in claim 26 (Final Act. 59–60).
Appellant argues that the rejection of claim 26 cannot be sustained
because although “Pestian refers to raw epilepsy diagnosis codes (see
Table 6 and paragraphs [0067] and [006[8]]), Pestian does not . . . mention
raw convulsion codes” and thus “does not disclose aggregating all raw
epilepsy diagnosis codes and all raw convulsion codes in the formatted
electronic medical records data of the patient into a higher level code in
a feature matrix using a medical classification scheme” (Appeal Br. 20).
Referencing Table 3, which appears at paragraph 41 of Appellant’s
Specification, Appellant maintains that this is so in that “Pestian only
references ICD-9 codes beginning with 345 (epilepsy diagnosis codes),
whereas the present application . . . discusses ICD-9 codes beginning with
780 – i.e., convulsion codes” (id.).
Responding to Appellant’s argument, the Examiner takes the position
that “Pestian teaches the argued limitation at paragraphs [0029]–[0032],”
i.e., that paragraphs 29–32 describe that the ontology for epilepsy comprises
the term class “‘seizure type,’ which under the broadest reasonable
interpretation includes convulsions” (Ans. 7). And the Examiner concludes,
“[t]herefore, the ICD-9 codes extracted as described in paragraph [0025], are
Appeal 2021-002263
Application 15/415,758
13
interpreted to read on both raw epilepsy codes as well as codes that are
related in the ontology of epilepsy which includes convulsions” (id.).
We find nothing, from our review of paragraph 25 of Pestian, which
discloses or suggests that the extracted ICD-9 codes include anything more
than the ICD-9-CM codes referenced in paragraphs 67 and 68 and Table 6.
And, to the extent the Examiner’s rejection is based on the view that Pestian
inherently discloses the argued limitation, more than speculation is required.
In particular, the Examiner must provide evidence and/or technical
reasoning, which is not present here, that make “clear that the missing
descriptive matter is necessarily present in the thing described in the
reference, and that it would be so recognized by persons of ordinary skill.”
Continental Can Co. USA v. Monsanto Co., 948 F.2d 1264, 1268 (Fed. Cir.
1991). “Inherency . . . may not be established by probabilities or
possibilities. The mere fact that a certain thing may result from a given set
of circumstances is not sufficient.” Id. at 1269 (quoting In re Oelrich,
666 F.2d 578, 581 (CCPA 1981)).
We are not persuaded, on the present record, that the Examiner
properly rejected claim 26 under 35 U.S.C. § 103. Therefore, we do not
sustain the Examiner’s rejection.
CONCLUSION
In summary:
Claim(s)
Rejected
35 U.S.C. § Reference(s)/Basis Affirmed Reversed
1–3, 8, 14 103 De Bruin,
Kheifetz,
Dilorenzo
1–3, 8, 14
Appeal 2021-002263
Application 15/415,758
14
Claim(s)
Rejected
35 U.S.C. § Reference(s)/Basis Affirmed Reversed
5, 9, 12, 13 103 De Bruin,
Kheifetz,
Dilorenzo,
Ebadollahi
5, 9, 12, 13
6, 7 103 De Bruin,
Kheifetz,
Dilorenzo, Pestian
6, 7
10, 11 103 De Bruin,
Kheifetz,
Dilorenzo, Benner
10, 11
15–17, 19,
20
103 De Bruin,
Kheifetz,
Dilorenzo, Narain
15–17, 19,
20
18 103 De Bruin,
Kheifetz,
Dilorenzo, Narain,
Miller
18
21, 22, 24,
25
103 De Bruin,
Kheifetz,
Dilorenzo, Narain,
Moon
21, 22, 24,
25
23 103 De Bruin,
Kheifetz,
Dilorenzo, Narain,
Moon, Miller
23
26 103 De Bruin,
Kheifetz,
Dilorenzo, Narain,
Pestian
26
Overall
Outcome
1–3, 5–26

REVERSED