Alzheimer's disease (AD) is the main cause of dementia for older people. Although several antidementia drugs such as donepezil, rivastigmine, galantamine, and memantine have been developed, the effectiveness of AD drug therapy is still far from satisfactory. Recently, the single nucleotide polymorphisms (SNPs) have been chosen as one of the personalized medicine markers. Many pharmacogenomics databases have been developed to provide comprehensive information by associating SNPs with drug responses, disease incidence, and genes that are critical in choosing personalized therapy. However, we found that some information from different sets of pharmacogenomics databases is not sufficient and this may limit the potential functions for pharmacogenomics. To address this problem, we used approximate string matching method and data mining approach to improve the searching of pharmacogenomics database. After computation, we can successfully identify more genes linked to AD and AD-related drugs than previous online searching. These improvements may help to improve the pharmacogenomics of AD for personalized medicine.

1. Introduction

Alzheimer’s disease (AD), the most common form of dementia, was first reported in 1906 [1]. In 2006, there were about 26.6 million AD patients worldwide and it was also common in southern Taiwan [2]. Although AD has been identified for a long time, most research progress was made in the recent 30 years [3]. However, no definitive cure is available for this disease and eventually it leads to death. Therefore, the drug discovery for Alzheimer’s disease remains challenging.

Single nucleotide polymorphisms (SNPs) are the most common variation in human genomes [4]. The importance of SNPs has been reviewed in genome-wide association studies for its association with disease susceptibility and drug metabolism [5, 6]. About 60–90% of the individual variation of drug response depends on pharmacogenomic factors. Therefore, SNP genotyping for candidate genes, pharmacological research, and drug discovery may play an increasingly important role in AD treatment. Meanwhile, increasing amounts of related information require the assistance of bioinformatics to construct the suitable databases and web servers.

Recently, PharmGKB (the Pharmacogenetics and Pharmacogenomics Knowledge Base) has been constructed to provide a comprehensive database for pharmacogenomic studies [7]. PharmGKB provides the pharmacogenetics research network in terms of SNP discovery and drug responses [8] with the fully curated knowledge for drug pathways, drug-related genes, and relationships among genes, drugs, and diseases. However, some information of different functions of PharmGKB is insufficient to allow convenient crosstalking between each other.

To solve this problem, we propose data mining method to improve the searching of pharmacogenomics of AD based on the download dataset of the PharmGKB resource.

2. Materials and Methods

The flowchart for pharmacogenomics in AD for personalized drug studies is shown in Figure 1. First of all, the AD-related drugs and genes are retrieved from PharmGKB download data using approximate string matching method and data mining approach. The genes associated with AD and the genes associated with a single Alzheimer’s drug are identified and compared with the online searching of PharmGKB. Then, numerous SNPs of genes associated with AD are identified. Through some SNP genotyping tools or assays, the association studies to AD-related drugs may be evaluated. Finally, the relevant information may be helpful for the personalized drug research.

2.1. AD-Related Drugs Using Approximate String Matching Based on PharmGKB Download Data

In order to study the pharmacogenomics of AD, we downloaded the known PharmGKB (the Pharmacogenetics and Pharmacogenomics Knowledge Base) (http://www.pharmgkb.org/downloads/) [9, 10] as source by the approximate string matching method [11] to find out all AD-related drug classes. The meaningful keywords associated with “Alzheimer’s disease” are shown in Table 1. Then, these found drug classes are used to find out associated genes by data mining approach. The description of the approximate string matching method for all AD-related drug classes gives a pattern string , that is, the meaningful keywords associated with “Alzheimer’s disease” and a text string , that is, the description for drug and disease retrieved from PharmGKB. Find a substring in that has the smallest edit distance [12] to the pattern . The pseudocode for the edit distance is shown in Algorithm 1.

(1)// initialization
(2)for to do
(4)end for
(5)for to do
(7)end for
(8)// edit distance
(9)for to do
(10) for to do
(11)  if( ( ) =  ( )) then
(13)  else
(14)   min   MIN[ ]
(15)     min + 1
(16)  end if
(17) end for
(18)end for

2.2. Data Mining Method for PharmGKB Download Data

In this study, we used a priori algorithm [13] for frequent item set mining and association rule learning over PharmGKB. The pseudocode for the a priori algorithm for data mining in PharmGKB is shown in Algorithm 2. At first, a priori algorithm has to find out the frequent gene in drug class for “Alzheimer’s disease.” A set of genes can be mined from each drug class. A priori algorithm is a “bottom up” approach, where frequent gene subsets are extended one item at a time (i.e., candidate generation) and groups of candidates are tested against the data. This algorithm is terminated when no further successful extensions are found.

(1)Apriori(PharmGKB, )
(2)  (frequent genes in drug class for Alzheimer’s disease)
(6) for each drug class PharmGKB
(8)  for each candidate gene
(9)   count gene  count gene  + 1
(10)   end for
(11) end for
(14)end while

2.3. SNP Searching for Genes Using the NCBI dbSNP

Every gene contains numerous SNPs. In order to find out SNPs of single gene for Alzheimer’s pharmacogenomics, NCBI dbSNP (http://www.ncbi.nlm.nih.gov/snp) is used to search in the study.

3. Results and Discussion

3.1. AD Information Based on PharmGKB Search

In PharmGKB online searching, the SNP variants, related genes, and drugs for AD are able to be retrieved. For example, the SNP information such as rs2066853 and rs6313 is provided (Figure 2). As shown in Figure 3, the AD-related genes such as ADRB1, AHR, HTR2A, MTHFR, and PTGS2 are identified and the related drugs such as olanzapine and risperidone are searched. This information may assist the researchers to study the pharmacogenomics of AD. Unfortunately, this PharmGKB online searching just provides limited information and it insufficiently copes with the complexity of the drug researches for Alzheimer’s personalized medicine.

3.2. PharmGKB-Based Data Mining of AD Information of Drug Classes or Gene Symbols

In current study, our proposed method is used to perform data mining for PharmGKB download data in terms of the keyword “Alzheimer’s disease.” As shown in Table 2, 22 kinds of AD-related drug classes are identified from “drug classes” of PharmGKB. Their corresponding PharmGKB accession ID, PubMed PMID, and the number of genes that are associated with AD-related drug classes are also presented. In total, 495 genes are identified for AD information of drug classes (see Supplementary file 1: gene information includes PharmGKB Accession Id, gene symbol, and publications are providing in different classes; it is available online at http://dx.doi.org/10.1155/2014/897653). Alternatively, 99 genes associated with AD are identified from “gene symbols” of PharmGKB in terms of the keyword “Alzheimer’s disease.” These results suggest that the same keyword, for example, Alzheimer’s disease, may identify different numbers of AD-associated genes between “drug classes” or “gene symbols” of PharmGKB.

After detailed examination, 67 genes in the gene symbols searching (bold fonts of gene names as shown in Table 3) are absent from the genes in the drug class searching (Table 2). Furthermore, genes corresponding to the drug “memantine” listed in Table 2 (drug classes) are not found in Table 3 (gene symbols). Therefore, some current drugs have identified a small number of AD-related genes in the drug class searching; however, the remaining AD-related genes that may affect AD-related drugs may be partly discovered in the gene symbols searching. These novelly identified AD-related genes may be the potential candidates for further drug development of AD. These results demonstrated that our proposed data mining method may be an improved AD pharmacogenomics study.

3.3. SNP Information of AD-Related Genes

The SNP statuses for 99 AD-related genes are also provided in Table 3. This SNP status for each gene is calculated from the online NCBI dbSNP queries. In general, many SNPs are found in these AD-related genes. Some SNPs of these genes have been reported to be associated with AD. For example, the APOE gene is found in Table 3 and one of its SNPs, such as ApoE epsilon 4 allele, has been reported to be associated with AD [14]. With suitable tools for SNP genotyping, these SNP candidates are warranted for the pharmacogenomics research of AD.

Currently, there are many high throughput SNP genotyping methods developed (as shown in Figure 1), including PCR resequencing [15], TaqMan probes [16], SNP microarrays [17], Matrix Assisted Laser Desorption/Ionization-Time of Flight (MALDI-TOF) [18], and others [19, 20]. Furthermore, some SNP genotyping tools or databases are also developed, such as SNP-RFLPing2 for comprehensive PCR-RFLP information based on SNPs [2124], algorithmic PCR-RFLP primer design and restriction enzymes for SNP genotyping [25, 26], and primer design for PCR-confronting two-pair primers (PCR-CTPP) [27, 28]. These tools and methods can provide useful and convenient information for SNP genotyping in the AD pharmacogenomics studies.

4. Conclusions

AD is the most common form of dementia for older people. The pharmacogenomics of AD still remains a challenge. In this study, we propose the pharmGKB-based data mining method to improve the gene discoveries for the potential AD-related drug candidates. With the assistance of bioinformatics, this improvement can help researchers to develop personal therapeutic drugs of AD.

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.


This work is partly supported by the National Science Council (NSC) in Taiwan under Grant nos. NSC101-2622-E151-027-CC3, NSC101-2221-E-464-001, NSC101-2320-B-037-049, NSC102-2221-E151-024-MY3, NSC102-2221-E214-039, and NSC102-2221-E-464-004, by the National Sun Yat-Sen University-KMU Joint Research Project (no. NSYSU-KMU 103-p014), and by the Ministry of Health and Welfare, Taiwan (MOHW103-TD-B-111-05).

Supplementary Materials

Gene information includes PharmGKB Accession Id, gene symbol, and publications are providing in different classes.

  1. Supplementary Tables