Bioagent Sample Matching using Elemental Composition Data: an Approach to Validation

Abstract

Sample matching is a fundamental capability that can have high probative value in a forensic context if proper validation studies are performed. In this report we discuss the potential utility of using the elemental composition of two bioagent samples to decide if they were produced in the same batch, or by the same process. Using guidance from the recent NRC study of bullet lead analysis and other sources, we develop a basic likelihood ratio framework for evaluating the evidentiary weight of elemental analysis data for sample matching. We define an objective metric for comparing two samples, and propose a method for constructing an unbiased population of test samples. We illustrate the basic methodology with some existing data on dry Bacillus thuringiensis preparations, and outline a comprehensive plan for experimental validation of this approach.

A validation framework for microbial forensic methods based on statistical pattern recognition

“…This includes the use of feature extraction techniques, choosing an appropriate metric, and how to do unbiased sampling of the relevant population. Much of this discussion is similar to that in earlier reports 1,2 . A particular metric, based on the knearest neighbor approach to pattern recognition is developed in an Appendix.…”

“…The smallest partial factorial design (a 2 x 2 version of Fig, 3) for sampling production methods generates 6 samples for each medium. For Bacillus anthracis, at least 25 media have been reported in the open and classified literature, and at least 12 of these are used commonly 1 . If only one analytical replicate of each sample were analyzed, this minimal experiment would still require 72 samples.…”

“…The "population" of possible growth and processing methods for a bacterial agent is virtual 1,2 , and can be represented by a unit process matrix, shown schematically in Figure 2, where each end-to-end process is broken down into unit process steps such as growth, separation of the microbe from the growth medium, washing, drying, milling, and combining with additives. For each unit process, there are a number of options, including the "null" option in which that particular unit process is not carried out.…”

“…In two previous studies 1,2 , it was suggested that validating the interpretation of microbial forensic analysis methods could be best accomplished within a framework that utilized ROC curves and estimates of likelihood ratios (ROC/LR). This framework was applied to the validation of sample matching, single hypothesis testing, and calibration in the context of current problems in microbial forensics.…”

Validation Strategies for Microbial Forensic Analysis

“…Therefore it is the intention of this paper to describe an approach to interpretive validation that can be applied to a wide variety of methods relevant to the microbial forensic discipline. An earlier study applied this framework narrowly to the problem of sample matching 25 .…”