Kinship Inferences for Second‐Degree Relatives With a Combination of STRs and Microhaplotypes

ABSTRACT

This study evaluated kinship identification using 19 short tandem repeats (STRs) from AGCU EX22 kit, 119 microhaplotypes (52 in Panel A + 67 in Panel B), and a theoretical 9622 genome‐wide single nucleotide polymorphisms (SNPs) panel (The 9K). The system power was assessed accounting for linkage on the basis of 198 Chinese Southern Han (CHS) family samples and simulated data. STR + Panels A and B distinguished full‐sibling (FS) from unrelated (un), with second‐degree effectiveness above 0.96 when log₁₀(LR) threshold was 4 and low false positive rate (<0.06%), but were insufficient for more distant kinship identification alone. The 9K panel was sufficient to separate third‐degree from un but not for fourth‐degree or more distant relatives. For discrimination among different kinships, STR + Panels A and B successfully distinguished all FS from half‐siblings (HS), uncle/aunt–nephew/niece (UN), and grandparents–grandson/granddaughter (GS) in real data, with minimal overlaps (<0.2%). Discriminating second‐ and third‐degree or within second‐degree relatives seems very challenging, even for 9K panel. Including a carefully selected additional relative (AR) may improve discrimination between GS and UN/HS, such as AR (GS‐un, which is GS for one person and un for the other) and AR (parent/child (PC)‐UN). Similarly, an AR (PC‐GS) may help distinguish UN from GS/HS, but there was slightly improved efficiency for HS.

This study shows that STR + microhaplotypes reliably distinguish close relatives, whereas a genome‐wide SNP panel improves detection of more distant kinship. However, distinguishing second‐ and third‐degree relationships remains challenging, though including additional relatives can enhance discrimination.

Abbreviations

ACR

allele coverage ratio

A _e

effective number of alleles

AR

additional relative

AUC

area under the curve

BWA

Burrows–Wheeler aligner

CHS

Chinese Southern Han

DOC

depth of coverage

FC

first cousin

FS

full‐sibling

FTL

false testing level

GATK

The Genome Analysis Toolkit

gDNA

genomic DNA

GS

grandparents–grandson/granddaughter

GUN

great uncle/aunt–nephew/niece

HS

half‐siblings

HUN

half‐uncle/aunt–nephew/niece

HWE

Hardy–Weinberg equilibrium

IBS

identical by state

IGG

investigative genetic genealogy

InDel

insertions/deletions

LD

linkage disequilibrium

MAF

minor allele frequency

MH

microhaplotypes

NGS

next‐generation sequencing

NPV

negative predictive value

PPV

positive predictive value

ROC

receiver operating characteristic

SE

standard error

SNP

single nucleotide polymorphism

STR

short tandem repeat

UN

uncle/aunt–nephew/niece

un

unrelated

1. Introduction

Kinship analysis plays a pivotal role in forensic genetics, supporting criminal investigations, resolving inheritance disputes, and identifying missing persons or disaster victims [1, 2, 3]. The advent of next‐generation sequencing (NGS) has expanded the repertoire of analyzable genetic markers [4, 5, 6]. This advancement is expected to enhance the efficiency of kinship analysis by expanding the variety of genetic markers used in forensic systems. Commonly used genetic markers include short tandem repeats (STRs), single nucleotide polymorphisms (SNPs), insertions/deletions (InDel), and microhaplotypes (MHs).

MHs were defined as genomic regions with a fragment length of 300 bp or less containing two or more closely linked SNPs [7]. MHs exhibit characteristics such as low mutation rates, lack of stutter artifacts, short amplicons, and non‐preferential amplification of each allele, making them advantageous for analyzing kinship relationships, especially in complex samples (e.g., degraded samples) [8, 9]. In recent years, a growing body of literature has reported NGS‐based panels and population data on MHs [10, 11, 12, 13, 14]. Du et al. [14] constructed a 188 MH panel that achieved ∼85% discrimination for second‐degree relatives at |t| = 4 threshold but proved inadequate for third‐degree relationships. Xue et al. [12] reported that the false testing level (FTL), the proportion of overlap between the likelihood ratio (LR) distributions of related and unrelated (un) pairs, was 0.11%, 8.99%, 35.46%, and 61.55% for second‐, third‐, fourth‐, and fifth‐degree relatives, respectively. The current MH panels remain insufficient for analyzing more distant kinship relationships. To enhance the analytical performance for such cases, it may be necessary to incorporate more MH loci or combine them with other genetic markers, such as STRs.

As kinship inference extends to more distant relationships, larger marker sets are required. The increase in marker quantity unavoidably introduces the impact of linkage. Linkage is a consequence of biological recombination, leading to closely located markers being inherited from parent to child at a higher frequency than independent markers [15]. Several strategies can be used in forensics to address linkage, such as (1) disregarding it, (2) excluding one locus in a linked marker pair, or (3) fully incorporating linkage [16]. Some studies have investigated the degree of impact when genetic linkage is neglected for a given marker set and case hypothesis [16, 17, 18, 19]. The previous research reported that “linkage has no effect at all in a pedigree unless (1) at least one individual in the pedigree (the central individual) is involved in at least two transmissions of genetic material, either as a parent or a child, and (2) that individual is a double heterozygote at the loci in question” [17]. Relationships, such as full‐sibling (FS), half‐sibling (HS), uncle/aunt–nephew/niece (UN), and grandparents–grandson/granddaughter (GS), contain double transmissions [18]. This study particularly focused on second‐degree relatives, thus fully incorporating linkage in our calculations.

Building on our previous 53‐MH (Panel A) and 67‐MH (Panel B) panels for Chinese Southern Han (CHS) population [11, 12], we combined them with 19 STRs (AGCU EX22) to evaluate the kinship efficiency for distinguishing first‐ to fifth‐degree relatives from un on the basis of 198 CHS family samples and simulation data. Next, this study aims to (1) distinguish second‐degree from FS and third‐degree relatives and (2) resolve internal second‐degree relationships (UN/GS/HS), with additional analyses examining performance improvement when incorporating additional relative (AR) data. Our study offers a valuable strategy and guidance applicable in forensic practice by combining STRs and MHs for complex kinship analyses.

2. Materials and Methods

2.1. Sample Preparation

This study analyzed 198 CHS samples from 14 unrelated families, including 133 previously sequenced individuals [11, 12], and 65 newly collected samples (kinship details in Figure S1). The dataset comprised 80 FS; 225 second‐degree duos, including 128 UN, 89 GS, and 8 HS; 131 third‐degree duos, including 97 first cousin (FC), 17 great uncle/aunt–nephew/niece (GUN), and 17 half‐uncle/aunt–nephew/niece (HUN); 70 fourth‐degree duos; 39 fifth‐degree duos; and 25 137 un duos. All participants or their legal guardians (for minors <16 years) provided informed consent, and sample collection was approved by the Ethics Committee of Sichuan University's Institute of Forensic Medicine (KS2022770).

2.2. Library Preparation and Sequencing Analysis

Genomic DNA (gDNA) from 65 samples was amplified for Panels A and B using the MultipSeq Custom Panel (iGeneTech Biotech, Beijing, China), with sequencing adapters added per the manufacturer's protocol. Libraries were sequenced on an Illumina NovaSeq 6000 system (PE150). Data processing followed our previously study [12]. The minimum depth of coverage (DOC) for each target region was set to 100 reads, with at least 30 reads per haplotype per locus. The average allele coverage ratio (ACR) and average DOC of each MH were assessed. All 198 samples were subjected to conventional STR typing using an AGCU EX22 kit (Applied ScienTech) [20].

2.3. Panels and Statistical Analysis

Panels A and B had 53 (average A _e = 7.43) and 67 MHs (average A _e = 5.34), respectively (detailed information in Table S1 and previous studies [11, 12]). Allele frequencies of STRs and MHs for CHS are shown in Table S2. The linkage disequilibrium (LD) was calculated using Arlequin v3.5.1.3 [21]. All markers were aligned with the Rutgers map [22], and genetic distances among these loci were calculated. For MHs, if no variant matched exactly, the nearest variant within 1000 bp up or downstream of the given locus was used as a substitute.

To explore the ability of SNPs based on the whole genome for second‐degree kinship analysis in CHS population, a theoretical SNP panel (9K panel) with 9622 loci was constructed. VCFtools [23] was used to filter candidate SNPs with a minor allele frequency (MAF) >0.01 on the basis of 105 un individuals from CHS in 1000 Genomes Project [24]. To ensure no LD between SNPs, PLINK2 [25] was adopted to establish the initial group with a correlation coefficient (r ²) <0.2 and higher A _e. Next, the SNPs with at least 150 kb separation in the same chromosome according to Kling and Tillmar [26] and with higher A _e were retained to form the final panel.

2.4. Kinship Analysis

The identical by state (IBS) score for each relationship was determined by summing the number of shared alleles between the two individuals for each marker. Merlin v1.1.2 software [27] incorporates the Lander–Green algorithm [15] and was adopted to evaluate the efficiency of the four combined panels (STR + Panel A, STR + Panel B, Panels A and B, and STR + Panels A and B) using pairwise kinship analysis. For an individual within a given family lineage, LR involves comparing the likelihood values (L) based on the genotypes of autosomal markers (G) under two alternative hypotheses: H₁ (the tested individual is a specific member of the familial lineage) and H₀ (the tested individual is un); that is, LR = L(G|H₁)/L(G|H₀). As in a previous study [28], LR_p,d was used to represent the LR that an individual would belong to one relationship, p, (hypothesis H_p) compared to another relationship, d, (hypothesis H_d). Specifically, LR_p,d = L(G|H_p)/L(G|H_d) = LR_p/LR_d, where LR_p or LR_d is the LR of H_p or H_d against H₀. The thresholds t ₁ and t ₂ were established as follows: (1) related: log₁₀(LR) > t ₁, (2) inconclusive: t ₂ ≤ log₁₀(LR) ≤ t ₁, and (3) un: log₁₀(LR) < t ₂. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), error rate, and inconclusive and system effectiveness were calculated for pairwise testing with H_p and H_d relationships. System effectiveness was defined as the proportion of individuals for whom the relationship was correctly classified. Simulations involving related and un cases for each kinship hypothesis were conducted 10 000 times using Merlin v1.1.2 software. Similar simulations and analyses were conducted for the 9K panel. Then, an AR was included in LR calculation to investigate the potentially increased discriminatory power within second‐degree relatives (UN, GS, and HS) on the basis of real and simulated data for STR + Panels A and B. We also simulated 50 loci with five alleles (A _e = 3.125) in each autosome and genetic distances of 1, 3, 5, 10, 20, 30, and 40 cM between loci 10 000 times in UN, GS, and HS.

3. Results

3.1. General Information of Panels and Sequencing Quality Assessment

All 65 samples were successfully genotyped for both panels. Panel A showed mean coverage of 4282 (range: 2475–5427) with ACR 0.91, whereas Panel B had 4106 (2213–6755) mean coverage and 0.90 ACR (Figures S2 and S3). No DOC‐ACR correlation was observed (p > 0.05), suggesting high sequencing efficiency. No mutations/recombinations were detected. Due to the allele number limitation in Merlin v1.1.2 (maximum 64 alleles per locus), mh20ZL‐003 from Panel A was excluded from analysis. Among STR markers, 19 loci successfully aligned with the Rutgers genetic map, necessitating exclusion of Penta E, Penta D, and Amelogenin. The final marker set's genetic distances are detailed in Table S3. LD analysis of the 119 MHs and 19 STRs revealed no significant locus pairing after Benjamini–Hochberg correction (Table S4).

3.2. Discrimination Between Relatives and Unrelated Individuals

The IBS‐based analysis of 138 markers showed progressively decreasing values with more distant relationships, with FS exhibiting the highest IBS scores and un pairs the lowest (Figure S4). The discrimination power of the four panels (STR + Panel A, STR + Panel B, Panels A and B, and STR + Panels A and B) was evaluated under the hypothesis that each duo was related (i.e., FS/second‐/third‐/fourth‐/fifth‐degree relatives) rather than un using log₁₀(LR) distributions on the basis of real and simulation data. As shown in Figure S5, STR + Panels A and B exhibited better performance than the others. System effectiveness, sensitivity, specificity, PPV, NPV, and error rate at different log₁₀(LR) threshold values were further assessed on the basis of real data in Table S5. The log₁₀(LR) values of FS showed no overlap with those of the un controls in either the real or simulation data. Among second‐degree, average log₁₀(LR) were 14.47 ± 0.52, 15.42 ± 0.63, and 15.93 ± 0.63 in UN, GS, and HS for STR + Panels A and B (0.04%, 0.06%, and 0.04% in FTL), respectively. The area under the curve (AUC) values of receiver operating characteristic (ROC) analysis were >0.999 for all panels (Figure 1A). For more distant relatives, STR + Panels A and B showed FTLs of 5.37%/28.57%/55.47% for third‐/fourth‐/fifth‐degree relatives, with log₁₀(LR) = 4 ± 0.26/1.32 ± 0.17/0.27 ± 0.11 (Figure S5) and AUCs = 0.997/0.937/0.803 (Figure 1B–D).

FIGURE 1.

The receiver operating characteristic curves for discrimination between relatives (second‐, third‐, fourth‐, and fifth‐degree) and the unrelated of four panels. The area under the curve (AUC) values of panels were presented. STR, short tandem repeat.

3.3. Discrimination Among Different Relationships

The discrimination between second‐degree relatives (UN/HS/GS) versus FS/FC was evaluated using real and simulated data (Figure 2, Table 1, Table S6). For second‐degree versus FS comparisons, real data showed clear separation with mean log₁₀(LR_p,d) values of 6.04 ± 0.15 (UN), 5.54 ± 0.58 (HS), and 6.41 ± 0.20 (GS) for second‐degree pairs versus −14.05 ± 0.55, −13.88 ± 0.54, and −13.69 ± 0.52 for FS pairs (Figure 2A). One UN pair involving twins was excluded (2–5 and 2–7 in Supplement Fig. 1.4 [12]). Simulation‐derived FTLs were minimal (0.17%–0.20%; Figure 2B). For second‐degree versus FC discrimination, mean log₁₀(LR_p,d) values were 1.96 ± 0.14 (UN), 2.22 ± 0.35 (HS), and 2.48 ± 0.20 (GS) versus −2.17 ± 0.14, −2.16 ± 0.13, and −2.32 ± 0.12 for FC (Figure 2C). At log₁₀(LR_p,d) ≥ 0, 10 pairs of 127 UN duos and 6 of 97 in the FC were incorrectly determined for the UN group, leading to a total error rate of 0.0714 (0.0699 in GS, 0.0476 in HS) in Table 1. With |t| = 1 threshold, effectiveness reached ∼0.84 for GS/HS but <0.8 for UN. Overlap rates between second‐degree and FC pairs were 11.26%–13.65% (Figure 2D).

FIGURE 2.

Discrimination between second‐degree relatives from full‐siblings (FS) and first cousin (FC). The scatter and box plot of likelihood ratio of one relationship versus others based on real data (A and C). The density distribution of likelihood ratio of one relationship versus others based on simulation data (B and D). A false testing level (FTL) was defined as the overlap (%) between the two datasets. GS, grandparents–grandson/granddaughter; HS, half‐siblings; UN, uncle/aunt–nephew/niece.

TABLE 1.

System powers of the STR + Panels A and B to discriminate between second‐degree relatives (uncle/aunt–nephew/niece [UN], grandparents–grandson/granddaughter [GS], and half‐siblings [HS]) from full‐siblings (FS) and first cousin (FC) in real data.

Relationship	t 1 (log10)	t 2 (log10)	Sensitivity	Specificity	PPV	NPV	Error rate	Inconclusive	Effectiveness
GS vs. FS	1	−1	1	1	1	1	0	0	1
	2	−2	1	1	1	1	0	0	1
	3	−3	0.95506	1	1	1	0	0.02367	0.97633
	4	−4	0.88764	1	1	1	0	0.05917	0.94083
UN vs. FS	1	−1	0.99213	1	1	1	0	0.00483	0.99517
	2	−2	0.98425	1	1	1	0	0.00966	0.99034
	3	−3	0.96850	1	1	1	0	0.01932	0.98068
	4	−4	0.92126	1	1	1	0	0.04831	0.95169
HS vs. FS	1	−1	1	1	1	1	0	0	1
	2	−2	1	1	1	1	0	0	1
	3	−3	0.87500	1	1	1	0	0.01136	0.98864
	4	−4	0.87500	1	1	1	0	0.01136	0.98864
GS vs. FC	0	0	0.89888	0.95876	0.95238	0.91176	0.06989	0.00000	0.93011
	1	−1	0.79775	0.88660	0.97260	0.97727	0.02151	0.13441	0.84409
	2	−2	0.60674	0.67010	0.98182	0.98485	0.01075	0.34946	0.63978
UN vs. FC	0	0	0.92126	0.93814	0.95122	0.90099	0.07143	0.00000	0.92857
	1	−1	0.74803	0.80412	0.98958	0.96296	0.01786	0.20982	0.77232
	2	−2	0.45669	0.59794	0.98305	0.98305	0.00893	0.47321	0.51786
HS vs. FC	0	0	1	0.94845	0.61538	1	0.04762	0	0.95238
	1	−1	1	0.83505	0.88889	1	0.00952	0.14286	0.84762
	2	−2	0.37500	0.59794	0.75000	1	0.00952	0.40952	0.58095

Abbreviations: NPV, negative predictive value; PPV, positive predictive value.

3.4. Discrimination Within Second‐Degree Relatives

The log₁₀(LR) values of L(G|H₁)/L(G|H₀) between UN, HS, and GS (H₁) and un (H₀) for real data and the predictive proportional distribution based on the highest likelihood of three hypotheses of UN, HS, and GS for STR + Panels A and B are presented in Figure 3. The panel correctly identified 74.01% of UN (94/127), 64.04% of GS (57/89), and 62.50% of HS (5/8) pairs (Figure 3A), with the average differences in log₁₀(LR) of pairwise analysis below 1. Simulation results showed lower accuracy: 55.34% (UN), 72.30% (GS), and 27.90% (HS) (Figure 3B). Analysis of simulated 50‐marker panels (A _e = 3.125) at varying genetic distances (1, 3, 5, 10, 20, 30, and 40 cM) revealed optimal discrimination at 5 cM, showing minimal overlaps: 4.62% (GS‐UN), 12.16% (GS‐HS), and 65.37% (UN‐HS) (Figure S6).

FIGURE 3.

The distribution of log₁₀(LR) between H₁ of uncle/aunt–nephew/niece (UN), grandparents–grandson/granddaughter (GS), and half‐sibling (HS) and H₀ of unrelated in real data (A), and the predictive proportional distribution based on the highest likelihood of three hypotheses of UN, HS, and GS in real data of STR + Panels A and B and simulated data of STR + Panels A and B and 9K panel (B). The three values in parentheses from left to right represent the proportions classified as UN, GS, and HS. STR, short tandem repeat.

3.5. An Additional Typed Relative for Discrimination Within Second‐Degree Relatives

We then examined the potential enhancement in the ability to identify UN, GS, and HS with a genotyped relative included in the LR calculation of the pairwise analysis and addressed the question of optimal selection for relatives. Details of the scenarios and hypotheses are shown in Figure 4 and Figures S7–S9. For GS pairs, log₁₀(LR_{GS, UN}) value under the hypotheses H_p = GS and H_d = UN increased in all cases when ①AR (FS‐GS) and ②AR (GS‐un) were added and was 98.58% in ⑥AR (parent–child (PC)‐UN) and 97.05% in ⑨AR (UN‐4th) (p < 0.05; Figure 4 GS/Figure S7A–E). The largest increment in log₁₀(LR_{GS, UN}) was found in ⑥AR (PC‐UN) (10.32 ± 0.48 in real data, p < 0.05), followed by ②AR (GS‐un) (8.82 ± 0.32) and ①AR (FS‐GS) (6.50 ± 0.83). The simulation results were consistent. Under H_d = HS, the increment was found for all cases in ②AR (GS‐un) with average log₁₀(LR_{GS, HS}) of 8.64 ± 0.32 for real data and 98.58% in ⑥AR (PC‐UN) with 10.64 ± 0.30 (p < 0.05; Figure 4 GS/Figure S7A–E). No significant improvement in the evidentiary power in situation ① for HS and ③④⑦ for both UN and HS.

FIGURE 4.

Investigation of the potential improvement of an additional relative (AR) in the evidentiary power within second‐degree relatives under the scenarios of grandparents–grandson/granddaughter (GS) ①–⑥, uncle/aunt–nephew/niece (UN) ①–⑥, and half‐sibling (HS) ①–②, respectively. For each scenario, the interested duos, that is, Person 1 (P₁) and Person 2 (P₂), and the added known relative of P₁ are colored in the family tree. Each AR was marked by his/her relationship with P₁ and P₂. For example, AR (FS‐GS) means that the first AR is the FS of P₁ and the GS of P₂, and the family tree was drawn in the first box for each scenario. The boxplots with lines connecting paired data points without and with ARs in real data were drawn on the left of the second box, and the bean plots for simulated data are on the right. FC, first cousin; FS, full‐siblings; GUN, great uncle/aunt–nephew/niece; HUN, half‐uncle/aunt–nephew/niece; PC, parent–child; un, unrelated.

For UN pairs, the log₁₀(LR_{UN, GS}) values were augmented in 99.24% cases in ⑤AR (PC‐FC), 96.37% in ⑥AR (PC‐GS), and 96.15% in ④AR (FS‐UN), and average values were 5.67 ± 0.17, 10.83 ± 0.42, and 10.41 ± 0.40, respectively (p < 0.05). The simulations showed similar results (Figure 4 UN/Figure S8A–D), and no significant enhancement in performance was observed in situation ①②③⑦. Under H_d = HS, only ⑥AR (PC‐GS) exhibited improvements in discrimination power with an average log₁₀(LR_{UN, HS}) of 20.68 ± 0.58 with real data and 21.17 ± 0.09 in the simulation (p < 0.05), and other situations showed no or little improvement (Figure 4 UN/Figure S8A–D). For HS duos, the small enhancement in efficiency was found in ①AR (FS‐HS) and ③AR (UN‐HUN) when H_d = UN and ①AR (FS‐HS) when H_d = GS (Figure 4 HS/Figure S9A,B).

3.6. Relationship Inference Based on the 9K Panel

After filtering, 9622 autosomal SNPs (mean A _e = 2.05) were selected for the 9K panel (detailed in Table S7 and Figure S10). The panel effectively separated second‐/third‐degree relatives from un individuals based on simulated data (effectiveness = 0.9983/0.8935 at log₁₀(LR) ≥ 2), though performance declined for distant relationships (fourth‐, sixth‐degree FTL = 0.04%–10.04%; Table S8, Figure S11). It perfectly distinguished FS from second‐degree relatives and achieved >95% accuracy for second‐degree versus FC pairs (Table S9). On the basis of the maximum likelihood of the three hypotheses within second‐degree relatives, GS identification was highly accurate (97.6%), whereas UN and HS showed reciprocal misclassification (∼25% each; Figure 3B). Pairwise analysis revealed moderate UN‐GS discrimination (FTL = 0.89%) but substantial UN‐HS overlap (FTL = 50.46%; Figure S12A). Poor results were found among the third‐degree relatives of FC, HUN, and GUN, with higher than 40% overlap in all pairs (Figure S12B).

4. Discussion

This study systematically evaluated four combined marker panels (19 STRs + 119 MHs) with full linkage incorporation in LR calculations, assessing kinship resolution using both empirical and simulated family data. Several factors can influence the calculation of LR, such as the number of loci used, the level of polymorphism of the loci in the population, the genetic distance between the linked loci, and the hypotheses constructed to assess the relationship between the tested individuals [16]. Linkage occurs within a pedigree, causing a specific haplotype to be inherited as a unit with higher probability, whereas LD is observed in a population with increased/decreased occurrence of certain haplotypes, which may be associated with recombination, genetic drift, and population migration [29, 30]. It is required for the Lander–Green algorithm to exclude situations where LD exists between genetic markers. Here, the Rutgers genetic map [22], a commonly referenced genetic map in forensic research for kinship analysis [5, 31], was used to estimate the genetic distance between markers.

For kinship analysis between second‐degree and un individuals, STR + Panels A and B exhibited higher system power. Our results outperform the study by Zhang et al. [5] and are slightly lower than those of Mo et al. [28], but we included the larger number of second‐degree relatives. For distant relatives, determining the kinship relationship can be challenging in the absence of other relatives, as they may not share any alleles at certain loci, whereas un individuals may coincidentally possess the same alleles at some loci. Recently, the use of genome‐wide high‐density SNP data to predict kinship relationships for more distant relationships has become a research hotspot, and investigative genetic genealogy (IGG) has emerged as a thriving field in forensic genetics [32, 33, 34]. To identify the optimal set of SNP loci in CHS population, we established a screening strategy to avoid LD and retain SNPs with higher A _e to construct a theoretical panel (9K panel). The results showed that this panel could separate third‐degree from un, with an effectiveness of 0.966, for fourth‐degree relatives when log₁₀(LR) = 4, and below 50% for more distant relatives. However, our selection process was relatively unsophisticated because it was not the focus of this study, and for practical applications, a more refined screening process is necessary.

For kinship analysis among different kinships, STR + Panels A and B could separate all FS (80 duos) from second‐degree (224 duos) in the real data with small overlaps (<0.2%). However, the panel was not sufficient to separate second‐ and third‐degree relatives. The integrated system of 56 STRs, 52 SNPs, and 72 InDels in Zhang et al. [5] showed similar results. Even for the 9K panel, the effectiveness could not reach 1 (about 0.96 in UN and HS, and 0.99 in GS when log₁₀(LR) = 2), with FTL values of ∼1%. Next, our analysis of second‐degree relatives (UN/HS/GS) revealed varying discriminatory power: GS showed strong differentiation from UN/HS, whereas distinguishing between HS and UN remained challenging, with ∼25% misclassification rates even using the 9K panel and >50% overlap in pairwise comparisons. The overestimated accuracy observed for HS in real samples is likely due to the small number of HS duos (n = 8) from limited pedigrees, whereas simulations may better reflect the true difficulty of distinguishing HS from other second‐degree relatives. The potentially increased power of adding a genotyped relative to the pairwise analysis of UN, GS, and HS was evaluated on the basis of real family samples and simulated data. Adding the AR of ②AR (GS‐un) and ⑥AR (PC‐UN) might have a significant improvement between GS and UN/HS, the same as ⑥AR (PC‐GS) for UN duos. A small increase in efficiency was observed when an AR was included for HS duos. Due to sample size limitations, the study could only cover certain possible scenarios of adding relatives, especially for HS. When LR calculations include an AR, there might be several scenarios for the alternative hypothesis, whereas only one scenario was considered for the preliminary survey to simplify the calculation (detailed in Figures S7–S9). This study only included autosomal markers; markers on the sex chromosome or mtDNA may offer additional information for kinship analysis [35, 36]. The restricted sample and family sizes in our study could potentially introduce bias, and thus further analyses with larger sample sizes are essential to validate the current results. In summary, this study provides a comprehensive exploration of the effectiveness of STR, MH, and SNP genetic marker panels in second‐degree relationship analysis, offering a valuable strategy and guidance for forensic practices.

Author Contributions

Jiaming Xue: conceptualization, project administration, data curation, methodology, software, visualization, writing – original draft, writing – review and editing. Qiushuo Wu: conceptualization, project administration, data curation, methodology, software, visualization, writing – original draft, writing – review and editing. Mengyu Tan: validation, methodology, visualization, software. Yazi Zheng: validation, methodology, visualization, software. Guihong Liu: validation, methodology, visualization, software. Haoyan Jiang: validation, investigation, software. Yuxuan Tan: validation, investigation, software. Yuanyuan Xiao: validation, investigation, software. Meli Lv: validation, investigation, software. Shengqiu Qu: funding acquisition, project administration, supervision, writing – review and editing. Weibo Liang: funding acquisition, project administration, supervision, writing – review and editing. Lin Zhang: funding acquisition, project administration, supervision, writing – review and editing.

Funding

This work was supported by the National Natural Science Foundation of China (Nos. 82271929 and 82402199).

Ethics Statement

The research protocol was reviewed and approved by the Ethics Committee at the Institute of Forensic Medicine, Sichuan University (approval reference number: KS2022770).

Consent

Informed consent was obtained from all study participants. In the case of individuals under 16 years of age, the legal guardian also provided their signature for informed consent.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: elps70122‐sup‐0001‐figureS1‐S12.pptx.

Supporting File 2: elps70122‐sup‐0002‐tableS1‐S9.xlsx.

Data Availability Statement

All data generated or analyzed during this study are included in the results of this article.