U.S. flag

An official website of the United States government, Department of Justice.

Success Story: Improving DNA Mixture Interpretation With the Help of Machine Learning

NCJ Number
252783
Date Published
March 2019
Agencies
NIJ-Sponsored
Publication Type
Factsheet
Grant Number(s)
2016-MU-BX-K110
Annotation
Following an overview of the current challenges facing DNA mixture analysis, this report describes the features and benefits of the Probabilistic Assessment for Contributor Estimation (PACETM), a machine learning method that improves DNA mixture interpretation.
Abstract
Previously, machine learning was not considered useful to forensic DNA interpretation, because of the lack of large and diverse data sets to "train" an algorithm; however, the rise of forensic databases and expansion of data-sharing practices has enabled machine learning to become a tool for addressing complex issues such as DNA mixture analysis. PACETM uses a hands-off approach for determining the number of contributors in traditional DNA data sets (those based on the highly variable regions in human DNA that allow for the identification of the contributor). It defines a local noise threshold for each region of the DNA profile analyzed and then applies machine learning algorithms in classifying the likely number of contributors in the sample. This tool is 20-percent more accurate than current methods of estimating the number of contributors in up to four-person mixtures, which is currently the upper limit. With an overall accuracy rate of 98 percent, PACE TM (patent pending) is able to improve the confidence in assessing the number of contributors. It is exclusively licensed to NicheVision and was supported through the U.S. Justice Department's National Institute of Justice. The goal is to have it operational in casework. It accommodates different DNA profiling kits and runs on standard computers.
Date Created: April 3, 2019