Development of an artificial intelligence-based automatic artifact identification tool for DNA profiling

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K18981
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 58040:Forensics medicine-related
• Research Institution: Kansai Medical University
• Principal Investigator: MANABE Sho 関西医科大学, 医学部, 助教 (00794661)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: DNA鑑定 / 法医学 / 人工知能 / アーチファクト / 混合試料

Outline of Final Research Achievements

In this study, we examined whether artifacts observed in forensic DNA testing can be identified by the random forest, which is one of the methods of machine learning. The data used were peaks (n = 43,158) obtained from DNA samples of a single-source profiles (n = 350) and mixed DNA samples from two to four individuals (n = 180). Three-fourths of all peaks were assigned to training data and one-fourth to test data, and machine learning was performed using the library "scikit-learn" for the Python programming language. The accuracy of both training and test data was 100% and approximately 98.9%, respectively. There was almost no difference in the accuracy for both data, and no obvious overfitting was observed.

Free Research Field

法遺伝学

Academic Significance and Societal Importance of the Research Achievements

DNA鑑定で扱われる試料は、複数人のDNAが混合した試料や量的に極めて少ない試料が多く、ヒトDNA由来のシグナルとアーチファクトを人の手で識別するのは困難である。本研究を通して、AIと既存のソフトウェアを組み合わせることで、人の手を介さなくても高精度でアーチファクトを判定できるようになった。もちろん、AIで100％正しく判定できるわけではないので、専門家によるレビューは欠かせないが、本研究成果は客観性が強く求められるDNA鑑定実務に大きく貢献できると考えられる。さらに、客観性の高いDNA鑑定が普及し、DNA鑑定の証拠能力が上がれば、犯罪立証だけでなく犯罪抑止にもつながるものと期待される。