Disaster Victim Identification in R

The purpose of dvir is to implement state-of-the-art algorithms for DNA-based disaster victim identification (DVI). In particular, dvir performs joint identification of multiple victims.

The methodology and algorithms of dvir are described in

• Egeland & Vigeland (2025): Kinship cases with partially specified hypotheses
• Vigeland & Egeland (2021): Joint DNA-based Disaster Victim Identification

The dvir package is part of the pedsuite, a collection of R packages for pedigree analysis. Much of the machinery behind dvir is imported from other pedsuite packages, especially pedtools for handling pedigree data, and forrel for the calculation of likelihood ratios. A comprehensive presentation of these packages, and much more, can be found in the book Pedigree Analysis in R.

Installation

To get the current official version of dvir, install from CRAN as follows:

install.packages("dvir")

Alternatively, the latest development version may be obtained from GitHub:

# install.packages("remotes")
remotes::install_github("magnusdv/dvir")

Joint DVI analysis

library(dvir)

We illustrate joint DVI analysis using a toy example from Vigeland & Egeland (2021), shown below. Three victim samples (V1, V2, V3) are to be matched against three missing persons (M1, M2, M3) from two families.

The hatched symbols indicate genotyped individuals. This simple example uses a single marker with 10 equifrequent alleles denoted 1, 2, …, 10. The observed genotypes are shown in the figure.

DNA profiles from victims are referred to as post mortem (PM) data, while the ante mortem (AM) data contain profiles from the reference individuals, R1 and R2.

The function dviJoint() performs a joint likelihood search over all valid assignments.

joint = dviJoint(example2)
#> DVI dataset: 
#>  3 victims (2M/1F): V1, V2, V3
#>  3 missing (2M/1F): M1, M2, M3
#>  2 reference individuals: R1, R2
#>  2 pedigrees: F1, F2
#> Markers: 1 
#> Number of assignments: 14 (exact)

The output contains the log-likelihood of each of the 14 assignments, and likelihood ratios (LRs) comparing the most likely solution with each of the others. Thus the first row has LR = 1, while larger values indicate less likely assignments:

	V1	V2	V3	loglik	LR	M1	M2	M3
1	M1	M2	M3	-16.118	1	V1	V2	V3
2	M1	M2	*	-17.728	5	V1	V2	*
3	*	M2	M3	-18.421	10	*	V2	V3
4	M1	*	M3	-20.030	50	V1	*	V3
5	*	M1	M3	-20.030	50	V2	*	V3
6	*	M2	*	-20.030	50	*	V2	*
7	*	*	M3	-20.030	50	*	*	V3
8	*	M1	*	-21.640	250	V2	*	*
9	M1	*	*	-21.640	250	V1	*	*
10	*	*	*	-21.640	250	*	*	*
11	M2	M1	M3	-Inf	Inf	V2	V1	V3
12	M2	M1	*	-Inf	Inf	V2	V1	*
13	M2	*	M3	-Inf	Inf	*	V1	V3
14	M2	*	*	-Inf	Inf	*	V1	*

The columns V1, V2 and V3 give the PM-oriented assignment: each victim is either paired with a missing person or remains unassigned ("*"). The columns M1, M2 and M3 give the AM-oriented view of the same assignments.

We see that the most likely joint solution is V1 = M1, V2 = M2 and V3 = M3.

Automated DVI workflow

For complete DVI analysis, the main function is dviSolve(). It combines several steps, including exclusions, undisputed identifications, pairwise LR calculations, joint analysis of unresolved families, and conclusions based on generalised likelihood ratios (GLRs).

We illustrate this with the built-in dataset fire, which includes three victim samples and three missing persons from the same family. The function plotDVI() gives an overview:

plotDVI(fire, style = 2)

Here is the complete workflow using LR threshold 10 000:

res = dviSolve(fire, detailedOutput = TRUE, threshold = 1e4, verbose = TRUE)
#> 
#> ------ Checking dataset  [+0.0s | 0.0s] ----------
#> 
#> Ok
#> 
#> ------ Nonidentifiable MPs  [+0.0s | 0.0s] -------
#> 
#> None 
#> 
#> ------ Exclusions, iteration 1  [+0.0s | 0.0s] ---
#> 
#> No exclusions
#> 
#> ------ Undisputed, iteration 1  [+0.0s | 0.0s] ---
#> 
#> No change; breaking loop
#> 
#> ------ AM-driven: Simple families  [+0.1s | 0.1s] 
#> 
#> 0 simple families
#> 
#> ------ AM-driven: Complex families  [+0.0s | 0.1s] 
#> 
#> 1 complex family: F1
#> 
#> Number of assignments: 34 (exact)
#> 
#>   Family Missing Sample       LR        GLR      Conclusion                            Comment
#> 1     F1      M1     V1 121.3476   70582.39     Match (GLR)                  Joint: {M1,M2,M3}
#> 2     F1      M2  V2/V3       NA 2947149.10 Symmetric match Full siblings: {M2, M3} = {V2, V3}
#> 3     F1      M3  V2/V3       NA 2947149.10 Symmetric match Full siblings: {M2, M3} = {V2, V3}
#> 
#> ------ PM-driven: Remaining victims  [+0.2s | 0.4s] 
#> 
#> No remaining victims
#> 
#> ------ Analysis complete  [+0.0s | 0.4s] ---------

The output contains AM-oriented and PM-oriented summary tables, including conclusions and comments. With detailedOutput = TRUE, the result also contains the pairwise LR matrix, the exclusion matrix, and the joint tables used in the analysis:

res$AM

Family	Missing	Sample	LR	GLR	Conclusion	Comment
F1	M1	V1	121.35	70582.39	Match (GLR)	Joint: {M1,M2,M3}
F1	M2	V2/V3	NA	2947149.10	Symmetric match	Full siblings: {M2, M3} = {V2, V3}
F1	M3	V2/V3	NA	2947149.10	Symmetric match	Full siblings: {M2, M3} = {V2, V3}

res$PM

Sample	Missing	Family	LR	GLR	Conclusion	Comment
V1	M1	F1	121.35	70582.39	Match (GLR)	Joint: {M1,M2,M3}
V2	M2/M3	F1	36.56	2947149.10	Symmetric match	Full siblings: {V2, V3} = {M2, M3}
V3	M2/M3	F1	1.31	2947149.10	Symmetric match	Full siblings: {V2, V3} = {M2, M3}

res$LRmatrix

	M1	M2	M3
V1	121.348	398.582	398.582
V2	0.583	36.564	36.564
V3	0.002	1.312	1.312

res$exclusionMatrix

	M1	M2	M3
V1	0	0	0
V2	0	0	0
V3	0	0	0

res$jointTables (truncated)

V1	V2	V3	loglik	LR	M1	M2	M3
M1	M2	M3	-257.73	1.00	V1	V2	V3
M1	M3	M2	-257.73	1.00	V1	V3	V2
*	M2	M3	-268.90	70582.39	*	V2	V3
*	M3	M2	-268.90	70582.39	*	V3	V2
M2	M1	M3	-272.63	2947149.10	V2	V1	V3
M3	M1	M2	-272.63	2947149.10	V2	V3	V1

Conclusion

In the fire example, V1 is identified as M1 by joint GLR analysis (Egeland & Vigeland, 2025). The two remaining victims are matched to the two missing full siblings M2 and M3, but the data do not distinguish between the two sibling assignments. This is reported as a symmetric match.