Change of Data-Driven Drug Design Trends Through Patent Analysis

Abstract

The pharmaceutical industry is one of the most research and development (R&D)-intensive industries. This industry has tried many strategies to overcome the limitations of a business model that had a high return and high risk. In recent years, the fourth industrial revolution has affected many industries, causing them to update their traditional production and business strategies to a “data science-based” approach. This data science methodology, based on the largely increased size of the data environment, has actively changed the pharmaceutical industry. Therefore, this study aimed to identify specific characteristics of data science innovation in the pharmaceutical industry through the analysis of patent data from the triadic patent databases from the United States, Japan, and Europe.

1. Introduction

The pharmaceutical industry is a highly research and development (R&D)-intensive sector. Since the 1970s, R&D activity in the pharmaceutical industry has increased rapidly. In addition to the intensive use of R&D, the challenges faced in pharmaceutical R&D have increased considerably. Consequently, developments in the pharmaceutical industry have taken place as a result of the increased difficulties encountered in the pharmaceutical R&D [1,2].

Pharmaceutical companies have attempted to find novel materials that are different from incumbent and traditional materials, such as small molecules and the so-called new molecular entities (NMEs), through the use of biologics and biological entities. Changes in the type of pharmaceutical R&D companies have also taken place; more recently, small companies have increased and, from 2004, these exceeded the productivity of the R&D departments of bigger companies [2].

In the early 2000s, pharmaceutical innovation was science-based, i.e., the innovation was highly dependent on scientific researchers, their network, and the collaborations among scientists and scientific institutes [3,4,5].

In the last 10 years, rapid changes in innovation have occurred as a result of improvements in data availability and computational ability. This innovation is a part of the “fourth industrial revolution,” which combines technologies and blends advanced services based on data science [6,7,8,9,10,11,12,13,14]. In detail, data science shows many possibilities of improving productivity and adopting new business models with promising technologies, such as wireless sensor networking, big data, artificial intelligence, cloud-based services, and so forth. Data science plays a role in enabling automation, optimization of production, data-driven innovation, variety of personalized services, and so forth in various industries [12,13].

The pharmaceutical industry has also used data sciences for innovation and attempted to overcome the drawbacks of business models that require tremendous amounts of budget for innovation. Hence, pharmaceutical innovators or companies have tried to find ways to reduce R&D cost and initiate less risky business models. Pharmaceutical innovation players have attempted to apply data-driven models to assist innovation in the pharmaceutical industry and achieve a lower failure rate in the drug approval process [9]. The pharmaceutical industry could attempt to develop a new business model supplying personalized services that are less risky and less profitable by applying data science with medical health information [6,7,12,14].

The pharmaceutical industry has been adopting data science-based innovation. These changes in pharmaceutical innovation can be identified through the examination of patent activities. The authors hypothesized that the technological innovation regime differed according to period, and the patent data could reflect the changes in the technological innovation regime. Thus, this study aimed to find innovation trends in the patent information that could provide a deeper and more specific understanding of data-science-based pharmaceutical innovation.

Section 2 of this paper highlights previous studies discussing trends in innovation in the pharmaceutical industries that have moved from science-based to a data science-based innovation. Section 3 of this paper explains the empirical methodology, showing the relationship between technology classification codes and the data retrieval process. Section 4 of this paper includes the descriptive statistics of data, empirical results, and a discussion on the implications of the empirical results. In Section 5 of the paper, the authors suggest conclusive results and offer an in-depth discussion on pharmaceutical innovation trends from the patent analysis.

2. Literature Review

Innovation is the process that refers to the development and application of a new product, process, or service, as assessed by the United States (US) Office of Technology [15]. Innovation is driven by social situations, and the trends can be monitored.

The pharmaceutical industry is under immense pressure to innovate. This pressure has increased exponentially owing to the vast increase in R&D assets and scientific and engineering personnel. The pharmaceutical industry has spent enormous amounts of money on R&D since the early 1960s. The extent of R&D expenditure was approximately twice that of all other industries in the 1980s, and it had increased by approximately three times that of all other industries by the late of 1990s, despite the decrease in the approval of new drug applications (NDAs) for new molecular entities (NMEs) and non-new molecular entities (non-NME) since 1996 [1].

Given the increasing complexity in pharmaceutical innovation, companies have attempted to find new ways to survive. The production of biopharmaceuticals compounds began in 1982. Between 1980 and 2004, the number of discovery projects in small companies slightly exceeded those in large companies. This resulted in a few small companies succeeding in their discovery projects, which, in turn, led to an increase in the mergers and acquisitions (M&A) of small companies [2].

In view of the increased computing ability and genomic knowledge, pharmaceutical innovation adopted science-based innovation in the early 2000s. The changes in innovation in the pharmaceutical industry have resulted in changes in the process of innovation and, subsequently, presented new possibilities. The key requirements for the successful management of science-based innovation in the pharmaceutical industry are a new international market strategy and internationalization and collaboration strategies in the R&D management [3]. The collaboration strategy should be used at the level of the national innovation system. Korea has a national innovation system using horizontal and vertical collaboration among government-sponsored research institutes to promote innovation within the pharmaceutical industry [4,5].

Since the late 2000s, most science has been impacted by the rapid increase in availability of information. As a result of this impact, traditional methodologies are re-adopted and re-applied with the largely increased size of available data. This phenomenon is known as data science, which has rebuilt traditional science based on this largely increased size of data. Data science has induced changes in many industries, as well as in science and academia. The healthcare industry has adopted data science to provide developed, qualified, and personalized services [6,7,12]. The strength of data science is that it merges various technologies; in particular, it combines pharmaceutical technology with digital and physical technologies [8,12].

Before the emergence of data science, science-based innovation was limited by productivity and dependent on previous research [16,17,18,19]. For example, in other R&D intensive industries, such as the energy industry, upstream companies insisted on the continued production of shale gas and oil from the early to the middle 2010s [20]. However, these companies needed to develop new production technologies by using data science [21,22]. Moreover, they attempted to change their business routine by horizontally and vertically expanding their business area [23].

In case of the pharmaceutical industry, some incumbent pharmaceutical innovators have applied modeling and simulating to reduce development costs. In addition, some companies have used data to perform a qualitative risk assessment [9].

However, recently, data-driven methods have emerged to reduce the high cost of extensive experimentation through the replacement of traditional simulation-based analysis with a quality by design (QbD) system [10]. The manufacturing system can be made more productive and efficient through the adoption of data science analysis methods. In the recent years, the continuous tablet manufacturing system is one of the most notable applications of data science-driven innovation [11]. The genomic data analysis-based development of drugs is used in clinical practice to reduce the cost of gathering the three billion human DNA [24].

A data science-based approach to the medical or pharmaceutical industry starts with the generation of data (electronic medical records (EMR) and electronic health records (HER)) related to medical, health, or clinical use and the transmission and storage of the data in the medical information system. The stored data are integrated and analyzed by a comprehensive clinical research approach dealing with data regarding genes, transcripts, proteins, and metabolites, known as omics. Interestingly, the omics-based services are also provided by traditional information technology companies as well as traditional medical care company [25].

The data science-based innovation or R&D begins with the measurement of highly utilizable data suitable for data science. Recent data science environments require high-performance hardware and time, to develop data science-based analytical models. However, for analysis and measurement, it is possible to operate even if the hardware is small and has a low performance capability. With the universalization of the wireless internet and the use of cloud-based services, small devices are now able to measure, analyze, store, and manage data and results for users, through network communications without the necessity for separate storage. In addition, with the popularization of the smartphone, most people have access to high-performance computing and wireless network equipment. Thus, the development of measuring equipment in such a technological environment, can be used as a means to secure competitiveness by lowering users’ entry barriers in healthcare services. Additionally, the development of multi-purpose health care measurement equipment has the advantage of applicability across various methods of service [13,14].

Personalized pharmaceutical services have also been developed by the application of 3-dimensional printing tools. The first approval of a 3D printed drug was announced in August 2015. As 3D-printed drugs are in the early stages of innovation, additional time will be required before they are suitable for clinical and general use by the customer. However, the 3D-printed drug technology has an enormous potential for innovation [26,27].

A role for data science in the pharmaceutical industry has been suggested in various forms within the integrated value chain. It begins with the generation of patient data, followed by the application of the generated and measured patient data in the conduction of large-scale clinical trials, and, finally, the production and distribution of personally optimized drugs based on the results of clinical research. The administration of an optimized drug generates clinical trial data, and the data subsequently obtained from the dose can be further utilized as clinical research, resulting in a virtuous cycle that improves the accuracy of the optimized drug [13,14].

3. Methodology

Many previous studies have provided micro evidence for the success of data science-driven pharmaceutical innovation; however, empirical and macro evidence is lacking. Therefore, to identify the empirical evidence of multidisciplinary innovation in the pharmaceutical industry, this study applied association rules and generated a map identifying the relationship between various fields of technology [28,29]. This study used the R project (version 3.4.3) with the packages ‘arules’ and ‘arulesViz’ to calculate association rules and visualize association maps [30,31,32].

3.1. Calculation of Association Criteria

Association rules show a meaningful and associated relationship through the calculation of the conditional probability among the items [33]. In this study, an item represents an International Patent Classification (IPC) code of some patent, and a transaction represents a patent of some transaction set. A transaction set is a set of granted patents for a particular analysis period. The authors divided the analysis period into several parts by taking into consideration the features of data set and results from previous studies. Details of the criteria for division of the analysis period are described in Section 3.3.

To obtain meaningful association rules and a relationship between technologies, it is necessary to select the correct evaluation criteria. There are three well-known criteria: support, confidence, and lift [28,29,33,34].

The concept of support is based on conditional probability. Support is defined as shown in Equation (1). The $i_x$ and $i_y$ of under every equation represents an item (an IPC code).

$$support\left(i_x\to i_y\right)=\frac{\mathrm{number}\mathrm{of}\mathrm{transactions}\mathrm{including}\mathrm{both}{i}_{x}and{i}_y}{totalnumberoftransactions}=P\left(i_x{\displaystyle \bigcap }{i}_y\right).$$

(1)

As the value of support of a certain transaction approaches 1, they (the items of the transaction) are considered to be more related. In other words, the greater the relative frequency is within the total transactions in a set, the closer the value is to 1.

The confidence of $i_x$ and $i_y$ can be calculated from Equation (2).

$$confidence\left(i_x\to i_y\right)=\frac{\mathrm{number}\mathrm{of}\mathrm{transactions}\mathrm{including}\mathrm{both}{i}_{x}and{i}_y}{totalnumberoftransactionsincluding{i}_x}=P\left(i_y|i_x\right)$$

(2)

This simple Equation (2) is a conditional probability. This simple equation is useful to understand the direction of the relationship between $i_x$ and $i_y$. The value of support, $support\left(i_x\to i_y\right)$ is the same as $support\left(i_y\to i_x\right)$. The difference in a pair of supports is understood as the size and direction of causal relationship [29].

The third criterion is lift. The lift can be calculated from the Equation (3).

$$lift\left(i_x\to i_y\right)=\frac{confidence\left(i_x\to i_y\right)}{support\left(i_y\right)}=\frac{P\left(i_y|i_x\right)}{P\left(i_y\right)}=\frac{P\left(i_x\bigcap i_y\right)}{P\left(i_x\right)P\left(i_y\right)}$$

(3)

As shown in Equation (3), the lift is calculated by applying confidence and support. The lift is not a type of probability. Therefore, it can converge to infinity. The value of lift represents the relationship of a transaction. If $lift\left(i_x\to i_y\right)$ is 1, $i_x$ and $i_y$ are independent of each other. If $lift\left(i_x\to i_y\right)$ is bigger than 1, $i_x$ and $i_y$ have a complementary relationship to each other. If $lift\left(i_x\to i_y\right)$ is smaller than 1, $i_x$ and $i_y$ have a substitutional relationship with each other [29].

3.2. Data

This study has retrieved a set of patent data from the KIPRIS (Korea Intellectual Property Rights Information Service) website [35]. The dataset was retrieved by search queries, including a combination of key words for three fields (title, abstract, and claim) and the technology classification codes for the IPC (International Patent Classification) field [21,36]. The search queries are summarized in

Table 1. Search queries.

Field	Contents	Operator
		Intra Field		Inter Field
IPC (International Patent Classification)	IPC = (A61B5/00+A61B6/50+A61B6/52+A61B8/52+G01S+G01S5/0278+G01V+G06F11/00+G06F16+G06F17/00+G06F17/20+G06F17/21+G06F17/2264+G06F17/27+G06F17/28+G06F17/30+G06F17/30002+G06F17/30047+G06F17/30067+G06F17/30076+G06F17/30153+G06F17/30194+G06F17/30268+G06F17/30289+G06F17/30312+G06F17/30318+G06F17/30386+G06F17/3061+G06F17/30734+G06F17/30861+G06F19/00+G06F19/18+G06F19/28+G06F19/30+G06F21/31+G06F21/50+G06F21/60+G06F21/6245+G06F3/01+G06F3/048+G06F9/00+G06K7/00+G06K7/1413+G06K9/00+G06Q+G08B21/00+G08B21/0205+G16B20+G16B50+G16C+G16H+G16Z99/00+H04L43/16+H04L9/32+H04M1/00+H04M1/725+H04N1/00+H04N21/00+H04N21/4135+H04N21/45+H04N5/232+H04Q9/00)	Or(+)		And(*)
Title	TL = [(Pharmaceutical+pharmacy+pharmacies)]	Or(+)	Or(+)
Abstract	AB = [(Pharmaceutical+pharmacy+pharmacies)]	Or(+)
Claim	CL = [(Pharmaceutical+pharmacy+pharmacies)]	Or(+)

.

As shown in Table 1, the queries have focused on pharmaceutical technology, including techniques related to data science. The combination of IPC codes refers to reports from KIPO (Korea Intellectual Property Office) that identify data technology IPC codes [37,38]. The retrieved data is summarized in

Table 2. Search results.

	China (CN)	Europe (EP)	Japan (JP)	United States (US)	Other Countries	Total
Applied patent count	460	486	300	4375	540	6161
(weight)	(7.5%)	(7.9%)	(4.9%)	(71.0%)	(8.8%)	(100.0%)
Rank of application	3	2	4	1
Granted patent count	16	131	118	1856	100	2221
(weight)	(0.7%)	(5.9%)	(5.3%)	(83.6%)	(4.5%)	(100.0%)
Rate of granted	3.5%	27.0%	39.3%	42.4%	18.5%	36.0%
Rank of granted	4	3	2	1

.

3.3. Descriptive Statistics

As shown in Table 2, the total number of applications between 1975 and 2018 was 6161. The country with the highest number of applications was the United States (US), with 4375 applications; the US was also the most successful country with a 42.4% of granted ratio. The US holds the largest number of granted patents in the dataset. China made the second-highest number of applications. In contrast to the, U.S.; China has a very low granted ratio (3.5%) and applied patent count (460).

In this study, the authors used patents from the popular triadic patent family of the United States, Europe, and Japan as the analysis sample [39]. Even for this popular triadic patent family, pharmaceutical-granted patent rate was low. The granted patent rate was 27.0%; thus, even the granted patent rate for the US did not exceed 50%. To avoid presenting results that contained insignificant information, disqualified patents were excluded. Thus, this study used only granted patent data. Finally, to reflect the characteristics of recent technology, the author restricted the sampling period.

As shown in

Table 3. Yearly granted patent counts by country [35].

	China (CN)	Europe (EP)	Japan (JP)	United States (US)	Other Countries	Total
1975	0	0	0	0	0	0
1976	0	0	0	0	0	0
1977	0	0	0	1	0	1
⋮	⋮
1997	0	1	1	12	2	16
1998	0	2	0	9	1	12
1999	0	6	0	14	4	24
2000	0	3	4	53	3	63
2001	1	12	3	66	6	88
2002	3	6	8	59	7	83
2003	3	13	5	87	8	116
2004	1	6	2	79	5	93
2005	3	7	7	91	8	116
2006	1	9	5	108	7	130
2007	1	6	7	137	2	153
2008	1	8	6	107	11	132
2009	1	4	12	118	7	142
2010	1	9	4	148	8	170
2011	0	6	8	114	5	133
2012	0	10	8	138	1	157
2013	0	3	10	136	2	151
2014	0	6	12	131	2	151
2015	0	6	5	105	0	116
2016	0	2	4	59	5	70
2017	0	0	3	42	0	45
2018	0	0	1	14	0	15
Total	16	131	118	1856	100	2221

, retrieved patent data counts rapidly increased in the year 2000. In the, U.S.; 14 patents were granted in 1999 and 53 patents were granted in 2000. Japan had zero granted patents in 1999 but a granted patent count in 2000. Europe slowly started to increase the patent count as compared with the US and Japan.

As shown in Figure 1, the rapid growth of yearly granted patent counts stopped after 2007, and has fluctuated until 2015. From 2016 to 2018, granted patent counts were dramatically decreased as the patent database system did not include and reflect the patent information for the previous three years. Thus, this study uses patent data from 2000 to 2015 for empirical analysis. The total number of granted patents from 2000 to 2015 was 1994, and the number of granted patents of the triadic countries, for the same period, was 1897. From 2000 to 2007, the total number of granted patents was 842, and the number of granted patents of the triadic countries was 783. From 2008 to 2015, the total number of granted patents was 1152, and the number of granted patents of triadic countries was 1142. The number of granted patents means the number of transactions of some set.

To provide a greater confidence on the breaking point of the innovation regime change, the authors checked the granted patent ratio for the dataset.

As shown in Figure 2, dramatic changes in the cumulative granted patent ratio can be easily identified by country. Since 2000, each country has shown different behaviors in cumulative granted patent ratio. The cumulative granted patent ratio in the US has decreased significantly. This may have resulted due to the rapidly increased activities in patent application. From 1975 to 1999, the cumulative granted patent count was 30. Subsequently, the US granted 31 patents in the following years. This could signal the change in the technological regime, from the active implementation of the innovation model, to a science-based pharmaceutical innovation. Thus, the focus of innovation activities progressed to the still developing science-based pharmaceutical innovations.

4. Results and Discussion

In this section, this study will present the association rules and maps representing the relationship of technologies for a specific innovation regime. The period between 2000 and 2007 was assumed to represent science-based innovation in the pharmaceutical industry. Data science-based innovation in the pharmaceutical industry has been represented by the period between 2008 and 2015. Moreover, this study has provided an in-depth discussion of the results from the perspective of the technological innovation theory.

This study set the minimum value of 0.01 for confidence and support. Therefore, the results showed only the association rules when the confidence and support exceed the minimum value.

4.1. Association Rule and Map of Science Based Pharmaceutical Technology

This section shows the association rules and map of the pharmaceutical patent data from 2000 to 2007. As shown in Figure 3, we found three big clusters with certain IPC codes as the centroid. The IPC codes located as the centroid were A61B5, G06Q50, and G06F19. First, Figure 3 showed some IPC codes surrounding A61B5 and the codes A61B8, A61B10, A61K49, A61K52, and G06Q50, which have a relationship with A61B5. Second, the IPC codes surrounding G06Q50, that is A61B5, G06Q10, Q06Q30, and G06Q40, which have a relationship with G06Q50, are shown in Figure 3. Third, the IPC codes surrounding G06F19, that is A61M5, C12Q1, G01N33, G06G7, G06Q30, and G06Q40, which have a relationship with G06F19, are shown in Figure 3. Detailed information of Figure 3 is shown in

Table A1. Relationships of clusters in the science-based pharmaceutical innovation period (2000–2007).

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
G06F19	A61B5	0.054	0.210	0.903	Substitutive	42
G06Q10		0.043	0.187	0.804	Substitutive	34
G06Q50		0.042	0.172	0.739	Substitutive	33
A61B8		0.041	1.000	4.302	Complementary	32
A61B10		0.040	1.000	4.302	Complementary	31
A61B10, A61B8		0.038	1.000	4.302	Complementary	30
G06F19, G06Q10		0.037	0.630	2.712	Complementary	29
G06Q10, G06Q50		0.034	0.290	1.249	Complementary	27
G06F19, G06Q50		0.033	0.765	3.290	Complementary	26
G06F19, G06Q10, G06Q50		0.032	0.926	3.984	Complementary	25
A61K49		0.015	0.923	3.971	Complementary	12
A61K51		0.010	0.800	3.442	Complementary	8
A61B5	G06F19	0.054	0.231	0.903	Substitutive	42
	G06Q50	0.042	0.181	0.739	Substitutive	33
	G06Q10	0.043	0.187	0.804	Substitutive	34
	A61B8	0.041	0.176	4.302	Complementary	32
	A61B10	0.040	0.170	4.302	Complementary	31

,

Table A2. Relationships of clusters in the science-based pharmaceutical innovation period (2000–2007).

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
G06Q10	G06F19	0.059	0.253	0.990	Substitutive	46
A61B5		0.054	0.231	0.903	Substitutive	42
G06Q50		0.043	0.177	0.693	Substitutive	34
A61B5, G06Q10		0.037	0.853	3.339	Complementary	29
G06Q10, G06Q50		0.034	0.290	1.137	Complementary	27
A61B5, G06Q50		0.033	0.788	3.085	Complementary	26
A61B5, G06Q10, G06Q50		0.032	0.926	3.625	Complementary	25
G01N33		0.023	0.692	2.710	Complementary	18
G06Q30		0.015	0.324	1.270	Complementary	12
A61M5		0.014	0.846	3.313	Complementary	11
G06Q40		0.013	0.192	0.753	Substitutive	10
G06G7		0.011	0.900	3.524	Complementary	9
C12Q1		0.010	0.615	2.409	Complementary	8
G06F19	G06Q10	0.059	0.230	0.990	Substitutive	46
	A61B5	0.054	0.210	0.903	Substitutive	42
	G06Q50	0.043	0.170	0.693	Substitutive	34

and

Table A3. Relationships of clusters in the science-based pharmaceutical innovation period (2000–2007).

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
G06Q10	G06Q50	0.119	0.511	2.084	Complementary	93
G06F19		0.043	0.170	0.693	Substitutive	34
A61B5		0.042	0.181	0.739	Substitutive	33
A61B5, G06Q10		0.034	0.794	3.239	Complementary	27
G06F19, G06Q10		0.034	0.587	2.394	Complementary	27
A61B5, G06F19		0.033	0.619	2.525	Complementary	26
A61B5, G06F19, G06Q10		0.032	0.862	3.516	Complementary	25
G06Q40		0.028	0.423	1.725	Complementary	22
G06Q30		0.014	0.297	1.212	Complementary	11
G06Q10, G06Q40		0.013	0.667	2.719	Complementary	10
G06Q50	G06Q10	0.119	0.484	2.084	Complementary	93
	G06F19	0.043	0.177	0.693	Substitutive	34
	A61B5	0.042	0.172	0.739	Substitutive	33
	G06Q40	0.028	0.115	1.725	Complementary	22

of Appendix A.

The IPC code located as the centroid has a relationship from some surrounding codes and a relationship towards some other surrounding codes. The feature of the centroid IPC code is that the relationship from surrounding IPC code to centroid IPC code has a higher confidence than the confidence of relationship from the centroid to the surrounding IPC code. This characteristic means that the technology corresponding to the centroid IPC code dispreads and is more commonly used when new patents or technology are invented. Furthermore, it would imply that the technology corresponding to the centroid IPC code is adopted, together with technologies corresponding to surrounding IPC code, but that the precedence of technology corresponding to the centroid IPC code would be an efficient way to invent technology and receive patents.

As another characteristic of the centroid IPC code, the centroid IPC code has a relationship with other centroid IPC codes. For example, the IPC code G06Q50 has a relationship with A61B5, and also the G06Q50 has a relationship with G06F19 through intermediate IPC codes, as shown in Figure 3. The code A61B5 has a relationship with G06Q50 and no relationship with G06F19. Actually, there is a direct relationship between the centroid IPC codes, but it is not weighty enough to be shown in Figure 3. The less weighty and directly connected relationships between the centroid IPC codes are presented in

Table 4. Various relationships of centroids of clusters in the science-based pharmaceutical innovation period (2000–2007).

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
Panel A: Single component and directly connected relationship between centroids of cluster
G06Q50	G06F19	0.059	0.177	0.693	Substitutive	46
G06F19	G06Q50	0.059	0.170	0.693	Substitutive	46
A61B5	G06F19	0.054	0.231	0.903	Substitutive	42
G06F19	A61B5	0.054	0.210	0.903	Substitutive	42
A61B5	G06Q50	0.042	0.181	0.739	Substitutive	42
G06Q50	A61B5	0.042	0.172	0.739	Substitutive	42
Panel B: Relationship from multi-components involving centroids of cluster to centroid of cluster
G06F19, G06Q10	A61B5	0.037	0.630	2.712	Complementary	29
G06Q10, G06Q50		0.034	0.290	1.249	Complementary	27
G06F19, G06Q50		0.033	0.765	3.290	Complementary	26
G06F19, G06Q10, G06Q50		0.032	0.926	3.984	Complementary	25
A61B5, G06Q10	G06F19	0.037	0.853	3.339	Complementary	29
G06Q10, G06Q50		0.034	0.290	1.137	Complementary	27
A61B5, G06Q50		0.033	0.788	3.085	Complementary	26
A61B5, G06Q10, G06Q50		0.032	0.926	3.625	Complementary	25
A61B5, G06Q10	G06Q50	0.034	0.794	3.239	Complementary	27
G06F19, G06Q10		0.034	0.587	2.394	Complementary	27
A61B5, G06F19		0.033	0.619	2.525	Complementary	26
A61B5, G06F19, G06Q10		0.032	0.862	3.516	Complementary	25
783 granted patents were used for analysis

.

Table 4 shows the relationships of G06Q10 with G06F19, A61B5 with G06F19, and A61B5 with G06Q50. The feature of panel A in Table 4 is that all relationships are substitutive, because every value of lift is less than one. However, the substitutive extent was different. The relationship between A61B5 and G06Q10 was 0.903. As that was close to one, most of the last substitutive relationships were in panel A of Table 4. G06Q50 has a relationship with G06F19 and A61B5. The relationships involving G06Q50 show a lower lift value than the relationship between A61B5 and G06F19. The lift value of the relationship between G06Q50 and G06F19 is 0.693, which is the most substitutive and lowest lift value in panel A of Table 4. The lift value of the relationship between A61B5 and G06Q50 was 0.739. Interestingly, the multi-component, which is the combination of centroid IPC codes and other IPC codes, always showed a complementary relationship toward the centroid IPC codes, as shown in panel B of Table 4. This had two implications. The first is that there was a need to develop combined technology in addition to single and independent technologies. The second is that the technologies of each cluster were developed simultaneously and were then developed independently by cluster. Moreover, the relationship from the triple component to the centroid of clusters showed the highest lift value, other than the relationship of multi-component cases, as shown in panel B of Table 4.

4.2. Association Rule and Map of Data-Science Based Pharmaceutical Technology

This section shows the association rules and map of pharmaceutical patent data from 2008 to 2015. There were two clusters encircled by different IPC codes, as shown in Figure 4. The IPC codes located as the centroid are A61B5 and G06F19. First, A61B5 is surrounded by A61K9, A61J3, A61N1, A61B6, A61M5, A61K38, and A61K51. Secondly, G06F19 is surrounded by C12Q1, A61J1, B65G1, B65B5, C07K14, G07F11, G06F7, A61K38, and A61M5. Detailed information of Figure 4 is shown in

Table A4. Relationships of clusters in the data science-based pharmaceutical innovation period (2008–2015).

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
A61K49	A61B5	0.050	0.966	5.026	Complementary	56
G06F19		0.036	0.106	0.549	Substitutive	40
A61B8		0.034	0.927	4.825	Complementary	38
A61B10		0.031	1.000	5.206	Complementary	35
A61B10, A61B8		0.029	1.000	5.206	Complementary	32
A61B8, A61K49		0.028	0.969	5.043	Complementary	31
A61B10, A61K49		0.027	1.000	5.206	Complementary	30
A61B10, A61B8, A61K49		0.027	1.000	5.206	Complementary	30
G01N33		0.027	0.462	2.403	Complementary	30
A61K31		0.025	0.500	2.603	Complementary	28
A61K51		0.019	0.955	4.969	Complementary	21
A61K9		0.019	0.656	3.416	Complementary	21
A61M5		0.018	0.526	2.740	Complementary	20
A61K49, G01N33		0.017	1.000	5.206	Complementary	19
A61N1		0.015	0.944	4.916	Complementary	17
A61K49, A61K51		0.014	1.000	5.206	Complementary	16
A61B6		0.013	0.833	4.338	Complementary	15
G06F19, G06Q50		0.013	0.138	0.716	Substitutive	15
A61B8, G01N33		0.013	1.000	5.206	Complementary	14
A61B8, A61K49, G01N33		0.013	1.000	5.206	Complementary	14
A61J3		0.013	0.452	2.351	Complementary	14
A61B10, G01N33		0.012	1.000	5.206	Complementary	13
A61B10, A61B8, G01N33		0.012	1.000	5.206	Complementary	13
A61B10, A61K49, G01N33		0.012	1.000	5.206	Complementary	13
A61B10, A61B8, A61K49, G01N33		0.012	1.000	5.206	Complementary	13
A61K38		0.011	0.444	2.314	Complementary	12
G06F19, G06Q10		0.011	0.143	0.744	Substitutive	12
A61B5	A61K49	0.050	0.262	5.026	Complementary	56
	G06F19	0.036	0.187	0.549	Substitutive	40
	A61B8	0.034	0.178	4.825	Complementary	38
	A61B10	0.031	0.164	5.206	Complementary	35
	G01N33	0.027	0.140	2.403	Complementary	30
	A61K31	0.025	0.131	2.603	Complementary	28

and

Table A5. Relationships of clusters in the data science-based pharmaceutical innovation period (2008–2015).

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
G06Q50	G06F19	0.098	0.308	0.905	Substitutive	109
G06Q10		0.075	0.327	0.961	Substitutive	84
G06F17		0.056	0.348	1.024	Complementary	62
G06Q10, G06Q50		0.054	0.397	1.168	Complementary	60
G07F17		0.048	0.885	2.602	Complementary	54
A61B5		0.036	0.187	0.549	Substitutive	40
A61J7		0.033	0.841	2.472	Complementary	37
G01N33		0.031	0.523	1.537	Complementary	34
G07F11		0.024	0.844	2.480	Complementary	27
G07F11, G07F17		0.023	0.867	2.547	Complementary	26
G06Q30		0.022	0.263	0.774	Substitutive	25
A61K31		0.021	0.411	1.207	Complementary	23
A61J1		0.020	0.786	2.309	Complementary	22
G06F7		0.019	0.525	1.543	Complementary	21
A61M5		0.017	0.500	1.470	Complementary	19
G06Q30, G06Q50		0.016	0.409	1.202	Complementary	18
G06F17, G07F17		0.015	0.895	2.630	Complementary	17
G06F17, G06Q50		0.015	0.586	1.723	Complementary	17
A61J7, G07F17		0.014	1.000	2.939	Complementary	16
G06Q50, G07F17		0.014	0.762	2.239	Complementary	16
B65B5		0.013	0.833	2.449	Complementary	15
A61B5, G06Q50		0.013	0.789	2.321	Complementary	15
G06F17, G06Q10		0.013	0.600	1.764	Complementary	15
G06Q10, G07F17		0.013	0.875	2.572	Complementary	14
A61K38		0.013	0.519	1.524	Complementary	14
B65G1		0.011	0.923	2.713	Complementary	12
A61B5, G06Q10		0.011	0.857	2.519	Complementary	12
C12Q1		0.011	0.750	2.204	Complementary	12
G06F17, G06Q10, G06Q50		0.011	0.667	1.960	Complementary	12
C07K14		0.011	0.632	1.856	Complementary	12
G06Q10, G06Q30		0.011	0.375	1.102	Complementary	12
G06F19	G06Q50	0.098	0.288	0.905	Substitutive	109
	G06Q10	0.075	0.222	0.961	Substitutive	84
	G06F17	0.056	0.164	1.024	Complementary	62
	G07F17	0.048	0.142	2.602	Complementary	54
	A61B5	0.036	0.106	0.549	Substitutive	40

of Appendix A.

Broadly, the IPC codes are divided into two group in Figure 4. The majority of IPC codes starting with A61 are linked with the A61B5 cluster and the majority of IPC codes starting with G06 are linked with the G06F19 cluster. In particular, there are intermediate IPC codes between A61B5 and G06F19, that is, A61K38 and A61M5.

The relationship between A61B5 and G06F19 is shown in

Table 5. Directly connected relationship between centroids of cluster for data science-based pharmaceutical innovation period (2008–2015) [35].

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
Panel A: Single component and directly connected relationship between centroids of cluster
A61B5	G06F19	0.036	0.187	0.549	Substitutive	40
G06F19	A61B5	0.036	0.106	0.549	Substitutive	40
Panel B: Relationship from multi component involving centroids of cluster to centroid of cluster
G06F19, G06Q50	A61B5	0.013	0.138	0.716	Substitutive	15
G06F19, G06Q10		0.011	0.143	0.744	Substitutive	12
A61B5, G06Q50	G06F19	0.013	0.789	2.321	Complementary	15
A61B5, G06Q10		0.011	0.857	2.519	Complementary	12
1114 granted patents were used for analysis

. The relationship between centroid cluster A61B5 and G06F19 was substitutive and the lift value of the relationship was 0.549. The interesting point was that the multi-component’s relationship was different according to object IPC code. The relationship toward G06F19 was complementary, but the relationship toward A61B5 was substitutive in panel B of Table 5. Every relationship from multi-component to centroid IPC code (A61B5) was complementary except the case involving G06F19, as shown in Table A4 of Appendix A. Moreover, every relationship from the multi-component to centroid IPC code (G06F19) was complementary, as shown in Table A5 of Appendix A. This implied that there was a need to develop combined technology compared with single and independent technologies when the purpose of developing the technology was represented by G06F19. Further, it implied that technology represented by A61B5 was developed as a precedent technology for the technology represented by G06F19.

Table 6. Relationships of centroids located between clusters A61B5 and G06F19 in the data science-based pharmaceutical innovation period (2008–2015) [35].

Centroid		Support	Confidence	Lift	Relationship	Count
From	To
A61M5	A61B5	0.018	0.526	2.740	Complementary	20
	G06F19	0.017	0.500	1.470	Complementary	19
A61K38	G06F19	0.013	0.519	1.524	Complementary	14
	A61B5	0.011	0.444	2.314	Complementary	12
1114 granted patents were used for analysis

shows the relationships between intermediate IPC codes (A61M5 and A61K38), and between A61B5 and G06F19. The lift value from A61M5 to A61B5 was 2.740, and the lift value from A61M5 to G06F19 was 1.470. The relationship from A61M5 to A61B5 was more complementary, higher than 1.270, than the relationship from A61M5 to G06F19. The lift value from A61K38 to A61B5 was 2.314, and the lift value from A61K38 to G06F19 was 1.524. The relationship from A61K38 to A61B5 was more complementary, higher than 0.79, compared to the relationship from A61K38 to G06F19. This implied that there was a need for the precedent development of technologies represented by A61M5 and A61K38.

4.3. Discussion

In this section, we will focus on the differences between the science-based period and the data science-based period to illuminate data-driven technological innovation in the pharmaceutical industry. The differences are summarized in

Table 7. Summary of differences between science-based innovation and data science-based innovation.

	Technological Innovation Regime
	Science-Based Innovation	Data science-Based Innovation
Panel A: Technological innovation characteristics
Purpose	Inventing new pharmaceutical entity	Inventing service or personalized, and qualified pharmaceutical entity
Strategy	Reducing failure in invention	Serviceable invention
Advantage	High return	Low risk
Disadvantage	High risk	Medium return
Panel B: Association rule and map
Centroid IPC code	A61B5, G06F19, G06Q50	A61B5, G06F19
Relationship among IPC code
Among centroid	Substitutive	Substitutive
From multi component to centroid	Complementary	Different according to centroid
Intermediate IPC code
Between A61B5 & G06F19	G06Q50	A61M5, A61K38
Between G06F19 & G06Q50	G06Q40, G06Q30
IPC code surrounding centroid
A61B5
Co-existing	A61K51	A61K51
Different	G06Q50, A61B8, A61B10, A61K49	A61M5, A61K38, A61K9, A61J3, A61N1, A61B6
G06F19
Co-existing	A61M5, C12Q1	A61M5, C12Q1
Different	G06G7, G06Q30, G06Q40, G01N33	A61K38, G06F7, G07F11, C07K14, B65B5, B65G1, A61J1
G06Q50	A61B5, G06Q10, G06Q40, G06Q30

.

First, the most remarkable difference existed in the intermediate IPC code between two periods. In the science-based innovation period, the centroid IPC codes were A61B5, G06Q50, and G06F19, and the intermediate IPC code between A61B5 and G06F19 was G06Q50. In the data science-based innovation period, the centroid IPC codes were A61B5 and G06F19, and the intermediate IPC code between A61B5 and G06F19 was A61M5 and A61K38. The most remarkable difference was that the intermediate technology changed from G06Q50 to A61K38 and A61M5.

A61M5 has no relationship with A61B5; it only has a relationship with G06F19 from 2000 to 2007. However, A61M5 has a complementary relationship with both A61B5 and G06F19. A61M5 represents “devices for bringing media into the body in a subcutaneous, intra-vascular or intramuscular way,” as shown in

Table A6. Description of IPC codes for the science-based innovation period (2000–2007).

IPC Code	Description
A61B5	Measuring for diagnostic purposes
A61B8	Diagnosis using ultrasonic, sonic or infrasonic waves
A61B10	Other methods or instruments for diagnosis
A61K49	Preparations for testing in vivo
A61K51	Preparations containing radioactive substances for use in therapy or testing in vivo
G06F19	Digital computing or data processing equipment or methods, specially adapted for specific applications
A61M5	Devices for bringing media into the body in a subcutaneous, intra-vascular or intramuscular way
C12Q1	Measuring or testing processes involving enzymes, nucleic acids or microorganisms; Compositions therefor; Processes of preparing such compositions
G01N33	Investigating or analyzing materials by specific methods
G06G7	Devices in which the computing operation is performed by varying electric or magnetic quantities
G06Q50	Systems or methods specially adapted for specific business sectors
G06Q10	Administration, Management
(Between G06F19 and G06Q50)
G06Q30	Commerce, e.g., shopping or e-commerce
G06Q40	Finance; Insurance; Tax strategies; Processing of corporate or income taxes

of Appendix A. This implied that the technology represented by A61M5 was more necessary and was invented during the development of clinical testing in the data science-based innovation period.

A61K38 represents “medical preparations containing peptides,” as shown in

Table A7. Description of IPC codes for the data science-based innovation period (2008–2015).

IPC Code	Description
A61B5	Measuring for diagnostic purposes
A61B6	Apparatus for radiation diagnosis
A61J3	Devices or methods specially adapted for bringing pharmaceutical products into particular physical or administering forms
A61K9	Medicinal preparations characterized by special physical form
A61K51	Preparations containing radioactive substances for use in therapy or testing in vivo
A61N1	Electrotherapy; Magnetotherapy; Radiation therapy; Ultrasound therapy
G06F19	Digital computing or data processing equipment or methods, specially adapted for specific applications
A61J1	Containers specially adapted for medical or pharmaceutical purposes
B65B5	Packaging individual articles in containers or receptacles
B65G1	Storing articles, individually or in orderly arrangements, in warehouses or magazines
C07K14	Peptides having more than 20 amino acids; Gastrin; Somatostatins; Melanotropins; Derivatives thereof
C12Q1	Measuring or testing processes involving enzymes, nucleic acids or microorganisms; Compositions thereof; Processes for preparing such compositions
G06F7	Methods or arrangements for processing data by operating upon the order or content of the data handled
G07F11	Coin-freed apparatus for dispensing, or the like, discrete articles
Intermediate
A61K38	Medicinal preparations containing peptides
A61M5	Devices for bringing media into the body in a subcutaneous, intra-vascular or intramuscular way

of Appendix A. The appearance of technology that used peptides in the data science-based innovation period may reflect the increase in the use of omics-based technologies [25]. In addition, we found similar cases showing directly connected and complementary relationships for digital computing (G06F19) and chemical materials (C07K14).

The direct relationship of A61B5 and G06F19 has changed. The lift value of A61B5 and G06F19 has decreased, and the difference was 0.354. This implied that the independent development of technology A61B5 and G06F19 occurred less frequently in the science-based innovation period than in the data science-based innovation period. Moreover, as described in Section 4.2, the precedence for development has been established. Thus, it may be implied that the relationship of technology has become more strict and concrete owing to the accumulation of technological development.

Differences also existed in surrounding IPC codes. A61J3 represented “devices or methods specially adapted for bringing pharmaceutical products into particular physical or administering forms.” The appearance of A61J3 may have indicated the invention for personalized drug forms, such as tablets printed by 3D printers [27]. The appearance of B65B5, representing “packaging individual articles in containers or receptacles,” and B65G1, representing “storing articles, individually or in orderly arrangement, in warehouses or magazines,” implied the occurrence of inventions for data processing technology dealing with personalized information [6,7,25] as shown in Table A7 of Appendix A.

Overall, the empirical results showed some agreement with the literature reviews. Specifically, some results indicated the invention of personalized and qualified services, and some results indicated the features of data science-based pharmaceutical characteristics in the discipline (omics). As summarized in Table 7, the suggested results and implications of data science-based pharmaceutical innovations are expected to bring about changes in the pharmaceutical industry to reduce risk and obtain medium return compared with the science-based innovation period.

We proposed an in-depth discussion about the most noteworthy IPC code ‘A61B5′ that refers to “measuring for diagnostic purposes.” The following interpretation of the observations of ‘A61B5′ were presented. Development of a measuring device may be relatively easier than the development of an analytical algorithm. Pharmaceutical companies have been competing through the development of measuring instruments. Traditional measuring equipment lacked sufficient potential to provide numerous and frequent measurements suitable for data science-based research. In addition, A61B5 may be included in patents to reflect the corresponding technical features as secondary or incidental technical components, rather than as a major component of the patent. Nevertheless, A61B5 has been included in many patents owing to the pharmacological industry’s data science-based innovation or business model generally including measurement and diagnostic technology as one of the technical features.

5. Conclusions

This paper presents an analysis of patent data distinguished by the period according to the technological innovation regime. The first period, that is science-based innovation, in which pharmaceutical innovation activity was based on simulation, focused on ways to find new NMEs, which involved high return but also high risk. During the second period, that is data science-based innovation, pharmaceutical innovation activity attempted to apply new ways to use data involving personal characteristics and information to identify services and products.

This study attempted to find macro evidence and trends in pharmaceutical innovation activity by using patent data. The empirical results characterized data science-based innovation technology and the points of accordance with the literature review. Despite these efforts to find macro trends, this study has limitations; thus, the accordance of empirical results with the literature review should be developed further to identify a more direct association.