AI detectors are poor western blot classifiers: a study of accuracy and predictive values.

Romain-Daniel Gosselin
Author Information
  1. Romain-Daniel Gosselin: Precision Medicine Unit, Biomedical Data Science Center, Lausanne University Hospital (CHUV), Lausanne, Switzerland. ORCID

Abstract

The recent rise of generative artificial intelligence (AI) capable of creating scientific images presents a challenge in the fight against academic fraud. This study evaluates the efficacy of three free web-based AI detectors in identifying AI-generated images of western blots, which is a very common technique in biology. We tested these detectors on AI-generated western blot images ( = 48, created using ChatGPT 4) and on authentic western blots ( = 48, from articles published before the rise of generative AI). Each detector returned a very different sensitivity (: 0.9583; : 0.1875; and : 0.7083) and specificity (: 0.5417; : 0.8750; and : 0.4167), and the predicted positive predictive value (PPV) for each was low. This suggests significant challenges in confidently determining image authenticity based solely on the current free AI detectors. Reducing the size of western blots reduced the sensitivity, increased the specificity, and did not markedly affect the accuracy of the three detectors, and only slightly improved the PPV of one detector (). These findings highlight the risks of relying on generic, freely available detectors that lack sufficient reliability, and demonstrate the urgent need for more robust detectors that are specifically trained on scientific contents such as western blot images.

Keywords

Associated Data

figshare | 10.6084/m9.figshare.26300464

References

  1. NPJ Digit Med. 2023 Apr 26;6(1):75 [PMID: 37100871]
  2. JCO Clin Cancer Inform. 2024 May;8:e2400077 [PMID: 38822755]
  3. J Am Acad Orthop Surg. 2024 Jun 27;: [PMID: 38941622]
  4. Ann Surg Oncol. 2024 Oct;31(10):6387-6393 [PMID: 38909113]
  5. Nature. 2024 Apr 8;: [PMID: 38589655]
  6. BMJ. 2022 Nov 28;379:e071517 [PMID: 36442874]
  7. Orthop Traumatol Surg Res. 2023 Dec;109(8):103694 [PMID: 37776949]
  8. Nature. 2023 Jun;618(7964):222-223 [PMID: 37258739]
  9. J Hematol Oncol. 2024 May 1;17(1):27 [PMID: 38693553]
  10. Orthop Traumatol Surg Res. 2023 Dec;109(8):103706 [PMID: 37838021]
  11. Biol Sport. 2023 Apr;40(2):615-622 [PMID: 37077800]
  12. J Biomed Inform. 2014 Apr;48:193-204 [PMID: 24582925]
  13. Cureus. 2023 May 19;15(5):e39224 [PMID: 37337487]
  14. Am J Obstet Gynecol. 2024 Aug;231(2):276.e1-276.e10 [PMID: 38710267]
  15. Patterns (N Y). 2023 Feb 28;4(3):100694 [PMID: 36960444]
  16. FEBS Lett. 2021 Jul;595(13):1751-1757 [PMID: 34180058]
  17. Curr Med Res Opin. 2024 Mar;40(3):353-358 [PMID: 38265047]
  18. Sci Rep. 2024 Nov 19;14(1):28677 [PMID: 39562595]
  19. Patterns (N Y). 2022 May 13;3(5):100509 [PMID: 35607625]
  20. Account Res. 2024 Mar 22;:1-17 [PMID: 38516933]
  21. Patterns (N Y). 2023 Mar 10;4(3):100706 [PMID: 36960451]

MeSH Term

Artificial Intelligence
Blotting, Western
Sensitivity and Specificity
Image Processing, Computer-Assisted
Predictive Value of Tests
Humans

Word Cloud

Created with Highcharts 10.0.0detectorsAIwestern:0imagesstudyblotsblotrisegenerativescientificthreefreeAI-generated=48detectorsensitivityspecificitypredictivePPVaccuracyResearchrecentartificialintelligencecapablecreatingpresentschallengefightacademicfraudevaluatesefficacyweb-basedidentifyingcommontechniquebiologytestedcreatedusingChatGPT4authenticarticlespublishedreturneddifferent958318757083541787504167predictedpositivevaluelowsuggestssignificantchallengesconfidentlydeterminingimageauthenticitybasedsolelycurrentReducingsizereducedincreasedmarkedlyaffectslightlyimprovedonefindingshighlightrisksrelyinggenericfreelyavailablelacksufficientreliabilitydemonstrateurgentneedrobustspecificallytrainedcontentspoorclassifiers:valuesdetectionAccuracyFraudImmunoblottingPapermillsethicsintegrity

Similar Articles

Cited By

No available data.