Combining artificial neural network classification with fully continuous probabilistic genotyping to remove the need for an analytical threshold and electropherogram reading

Standard processing of electrophoretic data within a forensic DNA laboratory is for one (or two) analysts to designate peaks as either artefactual or non-artefactual in a process commonly referred to as profile ‘reading’. Recently, FaSTR™ DNA has been developed to use artificial neural networks to automatically classify fluorescence within an electropherogram as baseline, allele, stutter or pull-up. These classifications are based on probabilities assigned to each timepoint (scan) within the electropherogram. Instead of using the probabilities to assign fluorescence into a category they can be used directly in the profile analysis. This has a number of advantages; increased objectivity in DNA profile processing, the removal for the need for analysts to read profiles, the removal for the need of an analytical threshold. Models within STRmix™ were extended to incorporate the peak label probabilities assigned by FaSTR™ DNA. The performance of the model extensions was tested on a DNA mixture dataset, comprising 2–4 person samples. This dataset was processed in a ‘standard’ manner using an analytical threshold of 50rfu, analyst peak designations and STRmix™ V2.9 models. The same dataset was then processed in an automated manner using no analytical threshold, no analysts reading the profile and using the STRmix™ models extended to incorporate peak label probabilities. Both datasets were compared to the known DNA donors and a set of non-donors. The result between the two processes was a very close performance, but with a large efficiency gain in the 0rfu process. Utilising peak label probabilities opens up the possibility for a range of workflow process efficiency gains, but beyond this allows full use of all data within an electropherogram.     

Maximizing intelligence obtained from higher-order DNA mixtures from volume crime samples

ABSTRACT

This work collates data from the analysis of complex mixtures analysed in STRmix during routine no-suspect volume crime work. It interrogates the upload rate for these types of mixtures and which component of the profile has been able to be interpreted for upload. The number of profiles giving multiple uploads and the amount of replicate PCR analysis has been collated.     

Investigating a common approach to DNA profile interpretation using probabilistic software

Recently there has been a drive for standardisation of DNA profile interpretation within and between different forensic laboratories. The continuous interpretation software STRmix⢢ has been adopted for use by laboratories in Australia and New Zealand for profile interpretation. Within this paper we examine the concordance in profile interpretation of three crime samples by twenty different analysts across twelve different international laboratories using STRmix⢢. The three profiles selected for this study exhibited a range of template and complexity. The use of probabilistic software has compelled a level of concordance between different analysts however there remain differences within profile interpretation, particularly with the objective assignment of the number of contributors to profiles.     

Probabilistic genotyping software: An overview

The interpretation of mixed profiles from DNA evidentiary material is one of the more challenging duties of the forensic scientist. Traditionally, analysts have used a “binary” approach to interpretation where inferred genotypes are either included or excluded from the mixture using a stochastic threshold and other biological parameters such as heterozygote balance, mixture ratio, and stutter ratios. As the sensitivity of STR multiplexes and capillary electrophoresis instrumentation improved over the past 25 years, coupled with the change in the type of evidence being submitted for analysis (from high quality and quantity (often single-source) stains to low quality and quantity (often mixed) “touch” samples), the complexity of DNA profile interpretation has equally increased. This review provides a historical perspective on the movement from binary methods of interpretation to probabilistic methods of interpretation. We describe the two approaches to probabilistic genotyping (semi-continuous and fully continuous) and address issues such as validation and court acceptance. Areas of future needs for probabilistic software are discussed.     

A fully continuous system of DNA profile evidence evaluation that can utilise STR profile data produced under different conditions within a single analysis

The introduction of probabilistic DNA interpretation systems has made it possible to evaluate many profiles that previously (under a manual interpretation system) were not. These probabilistic systems have been around for a number of years and it is becoming more common that their use within a laboratory has spanned at least one technology change. This may be a change in laboratory hardware, the DNA profiling kit used, or the manner in which the profile is generated. Up until this point, when replicates DNA profiles are generated, that span a technological change, the ability to utilise all the information in all replicates has been limited or non-existent. In this work we explain and derive the models required to evaluate (what we term) multi-kit analysis problems. We demonstrate the use of the multi-kit feature on a number of scenarios where such an analysis would be desired within a laboratory. Allowing the combination of profiling data that spans a technological change will further increase the amount of DNA profile information produced in a laboratory that can be evaluated.     

Interpreting a major component from a mixed DNA profile with an unknown number of minor contributors

Modern interpretation strategies typically require an assignment of the number of contributors (N) to a DNA profile. This can prove to be a difficult task, particularly when dealing with higher order mixtures or mixtures where one or more contributors have donated low amounts of DNA. Differences in the assigned N at interpretation can lead to differences in the likelihood ration (LR). If the number of contributors cannot reasonably be assigned, then an interpretation of the profile may not be able to be progressed.In this study, we investigate mixed DNA profiles of varying complexity and interpret them altering the assigned N. We assign LRs for true- and non- contributors and compare the results given different assignments of N over a range of mixture proportions. When a component of a mixture had a proportion of at least 10%, a ratio of at least 1.5:1 to the next highest component, and a DNA amount (as determined by STRmix™) of at least 50 rfu, the LR of the component for a true contributor was not significantly affected by varying N and was therefore suitable for interpretation and the assignment of an LR. LRs produced for minor contributors were found to vary significantly as the assigned N was changed. These heuristics may be used to identify profiles suitable for interpretation.     

Identifying common donors in DNA mixtures,with applications to database searches

Several methods exist to compute the likelihood ratio LR(M, g) evaluating the possible contribution of a person of interest with genotype g to a mixed trace M. In this paper we generalize this LR to a likelihood ratio LR(M₁, M₂) involving two possibly mixed traces M₁ and M₂, where the question is whether there is a donor in common to both traces. In case one of the traces is in fact a single genotype, then this likelihood ratio reduces to the usual LR(M, g). We explain how our method conceptually is a logical consequence of the fact that LR calculations of the form LR(M, g) can be equivalently regarded as a probabilistic deconvolution of the mixture.Based on simulated data, and using a semi-continuous mixture evaluation model, we derive ROC curves of our method applied to various types of mixtures. From these data we conclude that searches for a common donor are often feasible in the sense that a very small false positive rate can be combined with a high probability to detect a common donor if there is one. We also show how database searches comparing all traces to each other can be carried out efficiently, as illustrated by the application of the method to the mixed traces in the Dutch DNA database.     

NOCIt: A computational method to infer the number of contributors to DNA samples analyzed by STR genotyping

Repetitive sequences in the human genome called short tandem repeats (STRs) are used in human identification for forensic purposes. Interpretation of DNA profiles generated using STRs is often problematic because of uncertainty in the number of contributors to the sample. Existing methods to identify the number of contributors work on the number of peaks observed and/or allele frequencies. We have developed a computational method called NOCIt that calculates the a posteriori probability (APP) on the number of contributors. NOCIt works on single source calibration data consisting of known genotypes to compute the APP for an unknown sample. The method takes into account signal peak heights, population allele frequencies, allele dropout and stutter—a commonly occurring PCR artifact. We tested the performance of NOCIt using 278 experimental and 40 simulated DNA mixtures consisting of one to five contributors with total DNA mass from 0.016 to 0.25 ng. NOCIt correctly identified the number of contributors in 83% of the experimental samples and in 85% of the simulated mixtures, while the accuracy of the best pre-existing method to determine the number of contributors was 72% for the experimental samples and 73% for the simulated mixtures. Moreover, NOCIt calculated the APP for the true number of contributors to be at least 1% in 95% of the experimental samples and in all the simulated mixtures.     

Exploring how the LR of a POI in a target sample is impacted by awareness of the profile of the background derived from an area adjacent to the target sample

DNA unrelated to an action of interest (background DNA) is routinely collected when sampling an area for DNA that may have originated from an action of interest. Background DNA can add to the complexity of a recovered DNA profile and could impact the discrimination power when comparing it to the reference profile of a person of interest. Recent advances in probabilistic genotyping and the development of new tools, now allow for the comparison of multiple evidentiary profiles to query for a common DNA donor. Here, we explore the additional discrimination power that can be gained by having an awareness of the background DNA present on a surface prior to the deposition of target DNA. Samples with varying number of contributors and DNA quantities were generated on cleaned plastic pipes (where ground truth was known) and items used by occupants of a single household (where ground truth was not known). The background consisted of deposits made by hands (touch) while target deposits were both touch and saliva. Samples were collected from areas consisting of only the background (A), the target and the background directly beneath it (B), and the target and additional surrounding background (B+C). Samples B and B+C yielded similar DNA amounts when the target consisted of saliva, but when the target consisted of touch, significantly more DNA was recovered from B+C. Subsequently generated DNA profiles were interpreted using STRmix™ and DBLR™. The first approach involved no conditioning while the second approach involved conditioning on the reference profiles of the known background DNA donors. The third approach involved conditioning on one common DNA donor between A and B or A and B+C. The fourth and final approach involved conditioning on two common DNA donors between A and B or A and B+C. As more information was applied to the analysis, the greater the increase in the LR for the comparison of the target sample to the POI. Conditioning on two common donors between the target and the background provided almost the same amount of information as conditioning on the references of the known background DNA donors. This resulted in an increase in the LR that was over 10 orders of magnitude for known donors in the target sample. Here we have demonstrated the value in collecting additional background samples from an area adjacent to a targeted sample, and that this has the potential to improve discrimination power.     

Identification and characterization of novel DIP-STRs from whole-genome sequencing data

In an attempt to enhance forensic DNA mixture deconvolution several alternative DNA typing approaches have been developed. Among these, DIP-STR compound markers can resolve extremely unbalanced two-source DNA mixtures of same-or-opposite sex donors, up to a 1:1000 minor:major DNA ratio. A forensic set of 10 markers was validated for casework and a larger set of 23 DIP-STRs has proven suitable to biogeographic ancestry inference and for prenatal paternity testing. Yet, to promote the widespread use of this original approach, more markers and multiplex panels need to be developed.To this end, here we describe an extended set of forensic DIP-STRs identified using currently available whole-genome sequencing datasets. Complete lists of Indels and STRs were obtained from reported frequencies of genetic variants of 76,156 genomes. About 3000 identified DIP-STRs candidates were shorter than 200 bp and 500 showed high haplotype variability estimated using the genotypes of individuals homozygous for the DIP or the STR. Here, we present 23 additional DIP-STRs validated for sensitivity, specificity and Swiss population variability. Finally, a set of 30 markers comprising seven previously validated ones is proposed for the prospective development of a forensic DIP-STR multiplex panel.     

A phylogenetic approach for haplotype analysis of sequence data from complex mitochondrial mixtures

Massively parallel (next-generation) sequencing provides a powerful method to analyze DNA from many different sources, including degraded and trace samples. A common challenge, however, is that many forensic samples are often known or suspected mixtures of DNA from multiple individuals. Haploid lineage markers, such as mitochondrial (mt) DNA, are useful for analysis of mixtures because, unlike nuclear genetic markers, each individual contributes a single sequence to the mixture. Deconvolution of these mixtures into the constituent mitochondrial haplotypes is challenging as typical sequence read lengths are too short to reconstruct the distinct haplotypes completely. We present a powerful computational approach for determining the constituent haplotypes in massively parallel sequencing data from potentially mixed samples. At the heart of our approach is an expectation maximization based algorithm that co-estimates the overall mixture proportions and the source haplogroup for each read individually. This approach, implemented in the software package mixemt, correctly identifies haplogroups from mixed samples across a range of mixture proportions. Furthermore, our method can separate fragments in a mixed sample by the most likely originating contributor and generate reconstructions of the constituent haplotypes based on known patterns of mtDNA diversity.     

A comparative study for the isolation of exogenous trace DNA from fingernails

Often fingernails from a victim or suspect involved in a physical assault, such as murder or sexual assault, are submitted to crime laboratories for DNA testing of foreign/exogenous biological material; however, very few studies have been conducted comparing the effectiveness of different sampling methods on the removal of foreign/exogenous DNA while minimizing the fingernail endogenous DNA. In this study three different sampling methods (swabbing, PBS soak, and PrepFiler® lysis buffer soak) were compared in order to identify one that minimizes the amount of endogenous DNA removed and maximizes the amount of foreign/exogenous male DNA removed. The samples were processed using the Tecan HIDEVO150 robot in order to reduce analyst time and the DNA mixtures were interpreted using the probabilistic genotyping software STRmix™. For each sampling method the quantity of male DNA, the mixture proportions, the number of foreign/exogenous male alleles detected, the amount of DNA degradation, and the discrimination power via the likelihood ratio obtained for the foreign/exogenous male DNA donor were determined and compared. The PrepFiler® lysis buffer soak and swabbing sampling methods appear to be equally effective at removing foreign/exogenous DNA from fingernails; however, the lysis buffer soak sampling method extracts more female endogenous DNA from the fingernail and the female DNA is degraded. Marginally higher likelihood ratios were obtained for the swab samples versus the PrepFiler® lysis buffer soak samples; therefore, it was determined that the swabbing sampling method was the best sampling method for the recovery of foreign exogenous DNA from fingernails while minimizing the amount of endogenous DNA removed.     

Using a multi-head,convolutional neural network with data augmentation to improve electropherogram classification performance

DNA profiles are generated in forensic biology laboratories around the world. It is possible that these profiles are assessed by two independent people in order for the profiles to be ‘read’. Recent work has been carried out to develop a neural network model to classify fluorescence in a DNA profile electropherogram and potentially replace one, or both human readers. The ability to use neural networks for this function has been programmed into the software FaSTR™ DNA, which has been validated for use in at least one laboratory in Australia. The work that previously developed a neural network system had a number of limitations, specifically it was computer intensive, did not make the best use of available data, and consequently the performance of this model was sub-optimal in some conditions (particularly for low-intensity peaks). In the current work a new neural network model is developed that makes various improvements on the old model, by using convolutional layers, a multi-head architecture and data augmentation. Results indicate that an improved performance can be expected for low-intensity profiles.     

Statistical analysis of a large file of data from STIR profiles of British Caucasians to support forensic casework

Data from nearly 2500 British Caucasians, profiled using an STR quadruplex, have been analysed. The data came from several laboratories and represent samples from different geographical distributions. Analysis of the combined files shows that previous reports of failed independence tests were the results of sampling effects. A further convincing proof is given of the robustness of the statistical methods used to estimate evidential value in case-work. Comparisons between different samples show that regional effects between Scotland and the South of England have no importance from the forensic viewpoint.