CoChat
Sign Up
Cited by 107Machine Learning in BioinformaticsAlzheimer's disease research and treatmentsFractal and DNA sequence analysis

An Efficient Classifier for Alzheimer’s Disease Genes Identification

Open access
Xu, Lei, Liang, Guangmin, Liao, Changrui, Chen, Gin-Den, Chang, Chi-Chang·2018·Molecules
DOI 10.3390/molecules23123140PMC PMC6321377
Open this paper in CoChat — free accountOpen PDF

Abstract

Alzheimer’s disease (AD) is considered to one of 10 key diseases leading to death in humans. AD is considered the main cause of brain degeneration, and will lead to dementia. It is beneficial for affected patients to be diagnosed with the disease at an early stage so that efforts to manage the patient can begin as soon as possible. Most existing protocols diagnose AD by way of magnetic resonance imaging (MRI). However, because the size of the images produced is large, existing techniques that employ MRI technology are expensive and time-consuming to perform. With this in mind, in the current study, AD is predicted instead by the use of a support vector machine (SVM) method based on gene-coding protein sequence information. In our proposed method, the frequency of two consecutive amino acids is used to describe the sequence information. The accuracy of the proposed method for identifying AD is 85.7%, which is demonstrated by the obtained experimental results. The experimental results also show that the sequence information of gene-coding proteins can be used to predict AD.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper proposes an SVM classifier that predicts Alzheimer’s disease from gene-coding protein sequences using 400-dimensional dipeptide frequency features, reporting 85.7% accuracy. It also introduces a curated AD/non-AD protein dataset and compares the method with several alternative classifiers and a feature-selection approach.

Key findings

  • The proposed SVM method achieved 85.7% accuracy, 0.857 recall, 0.857 AUC, and 0.714 MCC for identifying AD from protein sequence information.
  • SVM performed better than random forest, naïve Bayes, LibD3C, AdaBoost, and Bayes network on accuracy, precision, recall, F-measure, and MCC, improving accuracy by nearly 1% to 7%; random forest had a higher AUC of 0.93.
  • The 400D dipeptide-frequency features outperformed the information theory feature method on accuracy, precision, F-measure, AUC, and MCC, though information theory had higher recall.
  • After CD-HIT filtering at 60% similarity, the benchmark dataset contained 279 AD-related proteins and 1,463 non-AD proteins.

Methods

  • Data were selected from UniProt; proteins with ambiguous residues were removed, and only proteins related to “Alzheimer’s disease” were kept. Initial selection left 310 AD and 312 non-AD proteins, then CD-HIT removed homologous sequences with more than 60% similarity, yielding 279 positive and 1,463 negative samples.
  • Each protein sequence was represented by a 400-dimensional vector encoding the frequency of two consecutive amino acids (20 × 20 possible pairs).
  • Classification used SVM with a radial kernel function via LIBSVM, evaluated by accuracy, precision, recall, F-measure, MCC, and AUC, and compared against random forest, naïve Bayes, LibD3C, AdaBoost, Bayes network, and an information theory feature method.

Limitations

  • The paper states that comparison with existing methods is difficult because most existing work uses the ADNI database and MRI information, so no direct benchmark comparison is provided.
  • The paper states future work will mine relationships between features and develop a webserver, implying the current model does not exploit feature relationships and is not yet deployed.
  • Evident from the text: the benchmark dataset is imbalanced (279 AD vs. 1,463 non-AD), and the paper does not report cross-validation or an independent test set, so the reported performance may be optimistic.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/molecules23123140

Cites

Statistics

Nothing to recompute.

Claims the paper makes

0 of 37 cited sources checked · 0 supported · 0 flagged

  • Findingcitation

    The AUC of the proposed method is 0.857 on the UniProt dataset.

    “The value of AUC for our method was 0.857 according to the UniProt dataset [41].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [41] Guo, 2014, Inuc-pseknc: A sequence-based predictor for predicting nucleosome positioning in genomes with pseudo k-tuple nucleotide composition, Bioinformatics source

      Not checked in the free run.

  • Findingown data

    SVM improves MCC by 0.8% to 20% over other classifiers.

    “Moreover, the value of MCC is improved by 0.8% to 20% by using SVM.”
    Quote in the paperSubstantial claim
  • Findingnone given

    After the selection process, there are 310 proteins related to AD and 312 non-AD proteins left in the benchmark dataset.

    “After the selection process, there are 310 proteins related to AD and 312 non-AD proteins left in the benchmark dataset.”
    Quote in the paperNot a substantial claim
  • Findingown data

    The MCC values for random forest and SVM reach a level of 0.7.

    “The values of MCC for random forest and SVM reach a level of 0.7.”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The benchmark dataset contains 279 AD-related proteins and 1,463 non-AD proteins after removing homologous sequences with >60% similarity.

    “As a result, a benchmark dataset with 279 proteins related to AD and 1,463 proteins not related to AD was used for the prediction model.”
    • Also stated in: § 3.1. Benchmark Dataset · plain
      “In other words, the benchmark dataset contains 279 positive samples in the positive subset (D+) and 1,463 negative samples in the negative subset (D−), respectively.”
    Quote in the paperSubstantial claim
  • The quote only partly states this claim. 3
  • May combine two claims. 8
  • Classified as probably not a substantial claim. 11

  • The quote only partly states this claim.

    Findingown data

    SVM improves performance by approximately 1% to 7.5% compared with other methods on precision, recall, and F-measure.

    “The performance is improved by SVM by approximately 1% to 7.5% as compared with in the case of the other methods.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    SVM performs better than the other classifiers on precision, recall, and F-measure.

    “The results are similar to those of Figure 3. SVM performs better than the other methods.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    SVM has MCC 0.714, which is better than the MCCs of the other mentioned methods.

    “Figure 8 shows that the MCC of SVM is 0.714, which is better than the MCCs of the other mentioned methods.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    The proposed SVM method using 400D dipeptide frequency achieves 85.7% accuracy in identifying AD.

    “The accuracy of the proposed method for identifying AD is 85.7%, which is demonstrated by the obtained experimental results.”
    • Also stated in: § 2.1. Results · plain
      “The accuracy of the proposed method was 85.7%, which means that the more than 85% of AD and non-AD samples were able to be classified correctly using the method in question.”
    • Also stated in: § 4. Conclusions · boosted
      “As demonstrated by the presented experimental results, the classification accuracy of our proposed method is 85.7%.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The 400D feature extraction method performs better than the information theory method on accuracy, precision, F-measure, AUC, and MCC.

    “Figure 2 shows that 400D performs better than information theory method on accuracy, precision, F-measure, AUC and MCC.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    SVM achieves accuracy 0.857, while AdaBoost, Bayes network, and naïve Bayes have accuracy about 0.8, LibD3C 0.84, and random forest 0.85.

    “The value of accuracy of AdaBoost, Bayes network, and naïve Bayes is about 0.8, while the accuracy of SVM is 0.857. The accuracy of LibD3C is 0.84. The accuracy of random forest is 0.85, which is comparative with that of SVM.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The recall of the proposed method is 0.857, meaning 85.7% of AD samples in the dataset were identified.

    “The recall of our method was 0.857, and the result shows that 85.7% of AD samples in the dataset could be identified in the experiment.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Random forest has AUC 0.93, which is better than other methods.

    “The value of AUC on random forest is 0.93, which is better than the values achieved via other methods.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    The key idea of SVM is that two groups are separated with a maximum margin by building a hyperplane.

    “Above all, the key idea of SVM is that two groups are separated with a maximum margin by building a hyperplane.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    There are, in total, 20 types of native amino acids.

    “There are, in total, 20 types of native amino acids.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning

    Since there are 20 native amino acids, the number of features of the combination of two consecutive amino acids is 400.

    “Since there are 20 native amino acids, the number of features of the combination of two consecutive amino acids is 400 (202).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning

    MRI scans are expensive and time-consuming because of the large size of the images.

    “However, MRI scans are expensive and the time required for scanning is significant because of the large size of the images.”
    • Also stated in: § Abstract · plain
      “However, because the size of the images produced is large, existing techniques that employ MRI technology are expensive and time-consuming to perform.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The principles of SVM were introduced in Chou and Cai, and more details are provided in Cristianini.

    “The principles of SVM were introduced in Chou and Cai [60,61], and more details are provided in Cristianini [62].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [60] Chou, 2002, Using functional domain composition and support vector machines for prediction of protein subcellular location, J. Biol. Chem. source

      Not checked in the free run.

    • cites [61] Cai, 2003, Support vector machines for predicting membrane protein types by using functional domain composition, Biophys. J. source

      Not checked in the free run.

    • cites [62] Cristianini, N., and Shawe-Taylor, J. (2000). An Introduction to Support Vector Machines: And Other Kernel-Based Learning Methods, Cambridge University Press. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    The metrics are used frequently in evaluating performance.

    “The metrics are used in evaluating the performance frequently [64,65,66,67,68,69,70,71,72].”
    Quote in the paperNot a substantial claim0 of 9 sources checked
    • cites [64] Chou, 2011, Some remarks on protein attribute prediction and pseudo amino acid composition, J. Theor. Biol. source

      Not checked in the free run.

    • cites [65] Chou, 2001, Using subsite coupling to predict signal peptides, Protein Eng. source

      Not checked in the free run.

    • cites [66] Lai, 2017, Sequence-based predictive modeling to identify cancerlectins, Oncotarget source

      Not checked in the free run.

    • cites [67] Su, R., Wu, H., Xu, B., Liu, X., and Wei, L. (2018). Developing a Multi-Dose Computational Model for Drug-induced Hepatotoxicity Prediction based on Toxicogenomics Data. IEEE/ACM Trans. Comput. Biol. Bioinform. source

      Not checked in the free run.

    • cites [68] Wei, 2017, Improved prediction of protein–protein interactions using novel negative samples, features, and an ensemble classifier, Artif. Intell. Med. source

      Not checked in the free run.

    • cites [69] Wei, 2018, Prediction of human protein subcellular localization using deep learning, J. Parallel Distrib. Comput. source

      Not checked in the free run.

    • cites [70] Wei, 2018, M6APred-EL: A Sequence-Based Predictor for Identifying N6-methyladenosine Sites Using Ensemble Learning, Mol. Ther. Nucleic Acids source

      Not checked in the free run.

    • cites [71] Liu, 2016, A systematic study on drug-response associated genes using baseline gene expressions of the cancer cell line encyclopedia, Sci. Rep. source

      Not checked in the free run.

    • cites [72] Xu, L., Liang, G., Shi, S., and Liao, C. (2018). SeqSVM: A Sequence-Based Support Vector Machine Method for Identifying Antioxidant Proteins. Int. J. Mol. Sci., 19. source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    It is difficult to compare the performance evaluation of the proposed method with existing methods.

    “Thus, it is difficult to compare the performance evaluation of our proposed method with the performance of existing methods.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning

    Soft SVM is proposed to tackle the problem of overfitting.

    “Soft SVM is proposed to tackle the problem.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    Proteins with ambiguous meanings were removed and only proteins related to Alzheimer's disease were kept to guarantee dataset validity.

    “To guarantee the validity of the dataset, the proteins with ambiguous meanings (such as “B”, “X”, and so on) is removed, and only the proteins related to “Alzheimer’s disease” are kept.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Most works in bioinformatics use SVM for classification.

    “Most works in bioinformatics [41,46,47,48,49,50,51,52,53,54,55,56,57,58,59] also use SVM for classification.”
    Quote in the paperNot a substantial claim0 of 10 sources checked
    • cites [41] Guo, 2014, Inuc-pseknc: A sequence-based predictor for predicting nucleosome positioning in genomes with pseudo k-tuple nucleotide composition, Bioinformatics source

      Not checked in the free run.

    • cites [46] Liu, B., Fang, L., Liu, F., Wang, X., Chen, J., and Chou, K.C. (2015). Identification of Real MicroRNA Precursors with a Pseudo Structure Status Composition Approach. PLoS ONE, 10. source

      Not checked in the free run.

    • cites [47] Feng, 2015, Identifying antioxidant proteins by using optimal dipeptide compositions, Interdiscip. Sci. Comput. Life Sci. source

      Not checked in the free run.

    • cites [48] Feng, P.M., Lin, H., and Chen, W. (2013). Identification of Antioxidants from Sequence Information Using Naïve Bayes. Comput. Math. Methods Med., 2013. source

      Not checked in the free run.

    • cites [49] Wei, L., Xing, P., Shi, G., Ji, Z.L., and Zou, Q. (2017). Fast prediction of methylation sites using sequence-based feature selection technique. IEEE/ACM Tran. Comput. Biol. Bioinform.

      Not checked in the free run.

    • cites [50] Zhang, 2018, Discriminating ramos and jurkat cells with image textures from diffraction imaging flow cytometry based on a support vector machine, Curr. Bioinform. source

      Not checked in the free run.

    • cites [51] Chen, W., Tang, H., Ye, J., Lin, H., and Chou, K.C. (2016). iRNA-PseU: Identifying RNA pseudouridine sites. Mol. Ther. Nucleic Acids, 5. source

      Not checked in the free run.

    • cites [52] Lin, 2014, Ipro54-pseknc: A sequence-based predictor for identifying sigma-54 promoters in prokaryote with pseudo k-tuple nucleotide composition, Nucleic Acids Res. source

      Not checked in the free run.

    • cites [53] Tseng, 2014, Application of machine learning to predict the recurrence-proneness for cervical cancer, Neural Comput. Appl. source

      Not checked in the free run.

    • cites [54] Tseng, 2017, Integration of data mining classification techniques and ensemble learning to identify risk factors and diagnose ovarian cancer recurrence, Artif. Intell. Med. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Premisenone given

    RBF is used widely because of its effectiveness and efficiency.

    “RBF is used widely because of its effectiveness and efficiency.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    SVM is widely used in classification problems because it performs better than some state-of-the-art supervised learning methods.

    “Since SVM performs better than some the-state-of-art supervised learning methods, SVM is widely used in classification problems.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: SVM performs better than some state-of-the-art supervised learning methods.
20 of 20

  • The paper relies on this without support.

    The proposed method is the first effort to identify AD by protein sequence information without MRI.

    “To the best of our knowledge, this study represents the first effort to identify AD by protein sequence information without the use of MRI.”

  • The paper relies on this without support.

    Identifying AD by using protein sequence information has not been widely done yet.

    “Identifying AD by way of using protein sequence information has not been widely done yet.”

  • The paper relies on this without support.

    SVM performs better than some state-of-the-art supervised learning methods.

    “Since SVM performs better than some the-state-of-art supervised learning methods, SVM is widely used in classification problems.”

  • The paper relies on this without support.

    Most existing diagnosis methods focus on identifying AD by MRI.

    “Most existing diagnosis methods focus on identifying AD by way of magnetic resonance imaging (MRI).”
4 of 4

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Negative sample count increases after redundancy removal

    Major

    The dataset is said to contain 312 non-AD proteins before homologous sequences are removed, but 1,463 afterward. Removing sequences cannot increase that count.

  • SVM is called best on all metrics despite a lower AUC

    Minor

    The overall conclusion says SVM performs better on the evaluated metrics, but the paper explicitly says random forest has the best AUC.

  • ADNI citation points to a DNA methylation paper

    Minor

    Reference [40], cited for the ADNI database, is listed as a paper about predicting DNA N4-methylcytosine sites.

  • Specificity description contradicts its equation

    Minor

    The text calls specificity a misclassification rate for AD proteins, while Equation (10) calculates the fraction of negative samples correctly classified.

  • UniProt dataset citation points to a nucleosome predictor

    Minor

    Reference [41], cited for the UniProt dataset, is listed as a paper about predicting nucleosome positions, not the UniProt database.

  • AUC is described as both an area and a ratio

    Minor

    The paper defines AUC as the area under the ROC curve, then calls it a ratio of specificity and sensitivity.

6 of 6

The citation lists haven’t been fetched yet

References 71 · Cited by 107

Research works printed but not linked: 1

  1. Liu, F., and Shen, C. (arXiv, 2014). Learning Deep Convolutional Features for MRI Based Alzheimer’s Disease Classification, arXiv.

Authors
Xu, L., Liang, G., Liao, C., Chen, G.-D., & Chang, C.-C.
Affiliations
  • Shenzhen Polytechnic University (CN)
  • Shenzhen University (CN)
  • Chung Shan Medical University Hospital (TW)
Corresponding author
  • Chen, Gin-Den
  • Chang, Chi-Chang
ORCID
Venue
Molecules
Publisher
MDPI AG
Published
2018-11-29
Volume
23
Issue
12
Pages
3140
ISSN
1420-3049 (ISSN-L)
DOI
10.3390/molecules23123140
PMC
PMC6321377
OpenAlex
W2902565488
Semantic Scholar
962f4a86dc62de51891ac9cb2be3192a44f4e48e
Cited by
107
References
71
Citations by year
2026: 2 · 2025: 5 · 2024: 6 · 2023: 10 · 2022: 22 · 2021: 24 · 2020: 30 · 2019: 8
FWCI
5.51
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecules (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Health Sciences › Medicine › Physiology › Alzheimer's disease research and treatments
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Fractal and DNA sequence analysis
Keywords
Alzheimer's disease, protein sequence features, support vector machine, disease prediction, early diagnosis, protein sequence classification
MeSH terms
Algorithms, Alzheimer Disease, Humans, Area Under Curve, Support Vector Machine

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review