Machine learning: a new era for cardiovascular pregnancy physiology and cardio-obstetrics research.

The maternal cardiovascular system undergoes functional and structural adaptations during pregnancy and postpartum to support increased metabolic demands of offspring and placental growth, labor, and delivery, as well as recovery from childbirth. Thus, pregnancy imposes physiological stress upon the maternal cardiovascular system, and in the absence of an appropriate response it imparts potential risks for cardiovascular complications and adverse outcomes. The proportion of pregnancy-related maternal deaths from cardiovascular events has been steadily increasing, contributing to high rates of maternal mortality. Despite advances in cardiovascular physiology research, there is still no comprehensive understanding of maternal cardiovascular adaptations in healthy pregnancies. Furthermore, current approaches for the prognosis of cardiovascular complications during pregnancy are limited. Machine learning (ML) offers new and effective tools for investigating mechanisms involved in pregnancy-related cardiovascular complications as well as the development of potential therapies. The main goal of this review is to summarize existing research that uses ML to understand mechanisms of cardiovascular physiology during pregnancy and develop prediction models for clinical application in pregnant patients. We also provide an overview of ML platforms that can be used to comprehensively understand cardiovascular adaptations to pregnancy and discuss the interpretability of ML outcomes, the consequences of model bias, and the importance of ethical consideration in ML use.

Identifying distant relatives using benchtop-scale sequencing.

The genetic component of forensic genetic genealogy (FGG) is an estimate of kinship, often conducted at genome scales between a great number of individuals. The promise of FGG is substantial: in concert with genealogical records and other nongenetic information, it can indirectly identify a person of interest. A downside of FGG is cost, as it is currently expensive and requires chemistries uncommon to forensic genetic laboratories (microarrays and high throughput sequencing). The more common benchtop sequencers can be coupled with a targeted PCR assay to conduct FGG, though such approaches have limited resolution for kinship. This study evaluates low-pass sequencing, an alternative strategy that is accessible to benchtop sequencers and can produce resolutions comparable to high-pass sequencing. Samples from a three-generation pedigree were augmented to include up to 7th degree relatives (using whole genome pedigree simulations) and the ability to recover the true kinship coefficient was assessed using algorithms qualitatively similar to those found in GEDmatch. We show that up to 7th degree relatives can be reliably inferred from 1 × whole genome sequencing obtainable from desktop sequencers.

Mixture detection with Demixtify.

The de facto genetic markers of forensics are short tandem repeats (STRs). There are many analytical tools designed to work with STRs, including techniques for analyzing and assessing DNA mixtures. In contrast, the nascent field of forensic genetic genealogy often relies on biallelic single nucleotide polymorphisms (SNPs). Tools designed for the forensic assessment of SNPs are somewhat lacking, especially for DNA mixtures. In this paper we introduce Demixtify, a program that detects DNA mixtures using biallelic SNPs. Demixtify is quite powerful; highly imbalanced mixtures can be detected (≤1:99, considering in silico and in vitro mixtures) when coverage is ample. Demixtify can also detect mixtures in low coverage (∼1×) samples (when the mixture is relatively balanced). Demixtify includes an empirical estimator of sequence error that is specific to the markers assayed, making it especially relevant to the forensic community. Orthogonal techniques are also developed to characterize in vitro mixtures, as well as samples thought to be single source, and the results of these approaches serve to validate the techniques presented.

MPKin-YSTR: Interpretation of Y chromosome STR haplotypes for missing persons cases.

Y chromosome Short Tandem Repeat (STR) haplotypes have been used in assisting forensic investigations primarily for identification and male lineage determination. The current SWGDAM interpretation guidelines for Y-STR typing provide helpful guidance on those purposes but do not address the issue of kinship analysis with Y-STR haplotypes. Because of the high mutation rate of Y-STRs, there are complex missing person cases in which inconsistent Y-STR haplotypes between true paternal lineage relatives will arise and cases with two or more male references in the same lineage and yet differ in their haplotypes. Therefore, more useful methods are needed for interpreting the Y-STR haplotype data. Computational methods and interpretation guidelines have been developed specifically addressing this issue, either using a mismatch-based counting method or a pedigree likelihood ratio method. In this study, a software program, MPKin-YSTR, was developed by implementing those more sophisticated methods. This software should be able to improve the interpretation of complex cases with Y-STR haplotype evidence. Thus, more biological evidence will be interpreted, which in turn will result in more investigation leads to help solve crimes.

Using unique molecular identifiers to improve allele calling in low-template mixtures.

PCR artifacts are an ever-present challenge in sequencing applications. These artifacts can seriously limit the analysis and interpretation of low-template samples and mixtures, especially with respect to a minor contributor. In medicine, molecular barcoding techniques have been employed to decrease the impact of PCR error and to allow the examination of low-abundance somatic variation. In principle, it should be possible to apply the same techniques to the forensic analysis of mixtures. To that end, several short tandem repeat loci were selected for targeted sequencing, and a bioinformatic pipeline for analyzing the sequence data was developed. The pipeline notes the relevant unique molecular identifiers (UMIs) attached to each read and, using machine learning, filters the noise products out of the set of potential alleles. To evaluate this pipeline, DNA from pairs of individuals were mixed at different ratios (1-1, 1-9) and sequenced with different starting amounts of DNA (10, 1 and 0.1 ng). Naïvely using the information in the molecular barcodes led to increased performance, with the machine learning resulting in an additional benefit. In concrete terms, using the UMI data results in less noise for a given amount of drop out. For instance, if thresholds are selected that filter out a quarter of the true alleles, using read counts accepts 2381 noise alleles and using raw UMI counts accepts 1726 noise alleles, while the machine learning approach only accepts 307.

Optimized variant calling for estimating kinship.

One of the fundamental goals of forensic genetics is sample attribution, i.e., whether an item of evidence can be associated with some person or persons. The most common scenario involves a direct comparison, e.g., between DNA profiles from an evidentiary item and a sample collected from a person of interest. Less common is an indirect comparison in which kinship is used to potentially identify the source of the evidence. Because of the sheer amount of information lost in the hereditary process for comparison purposes, sampling a limited set of loci may not provide enough resolution to accurately resolve a relationship. Instead, whole genome techniques can sample the entirety of the genome or a sufficiently large portion of the genome and as such they may effect better relationship determinations. While relatively common in other areas of study, whole genome techniques have only begun to be explored in the forensic sciences. As such, bioinformatic pipelines are introduced for estimating kinship by massively parallel sequencing of whole genomes using approaches adapted from the medical and population genomic literature. The pipelines are designed to characterize a person's entire genome, not just some set of targeted markers. Two different variant callers are considered, contrasting a classical variant calling algorithm (BCFtools) to a more modern deep convolution neural network (DeepVariant). Two different bioinformatic pipelines specific to each variant caller are introduced and evaluated in a titration series. Filters and thresholds are then optimized specifically for the purposes of estimating kinship as determined by the KING-robust algorithm. With the appropriate filtering and thresholds in place both tools perform similarly, with DeepVariant tending to produce more accurate genotypes, though the resultant types of inaccuracies tended to produce slightly less accurate overall estimates of relatedness.

A genotype likelihood function for DNA mixtures.

The recent advent of genetic genealogy has brought about a renewed interest in genome-scale forensic analyses, of which kinship estimation is a critical component. Most genomic kinship estimators consider SNPs (single nucleotide polymorphisms), often leveraging the co-inheritance of shared alleles to inform their analyses. While current estimators cannot directly evaluate mixed samples, there exist well-established SNP-based kinship estimators tailored to considering challenged samples, including low-pass whole genome sequencing. As an example, several studies have shown remarkable success in imputing genotype posterior probabilities in low template samples when linked sites are considered. Critical to these approaches is the ability to account for genotype uncertainty; the lack of an expression for a genotype likelihood in imbalanced mixtures has prevented direct application. This work develops such an expression. The formulation is fully compatible with genotype imputation software, suggesting a genomic pipeline that estimates genotype likelihoods, performs imputation, and then estimates kinship when the sample is a mixture. Further, when framed as an imbalanced mixture, the problem of mixture deconvolution is reducible to the problem of genotyping mixed samples. Herein, the ability to genotype two-person mixtures is assessed through example and in silico settings. While certain mixture scenarios and classes of sites are inherently inseparable, simulations of read depths between 60 and 190 appear to produce likelihoods of sufficient magnitude to deconvolve two-person mixtures whenever the mixture fraction is moderately imbalanced. The described approach and results suggest a path forward for estimating the kinship coefficient (and similar inferences on relatedness) when the sample is a mixture.

Techniques for estimating genetically variable peptides and semi-continuous likelihoods from massively parallel sequencing data.

Forensic genetic investigations typically rely on analysis of DNA for attribution purposes. There are times, however, when the amount and/or the quality of the DNA is limited, and thus little or no information can be obtained regarding the source of the sample. An alternative biochemical target that also contains genetic signatures is protein. One class of genetic signatures is protein polymorphisms that are a direct consequence of simple/single/short nucleotide polymorphisms (SNPs) in DNA. However, to interpret protein polymorphisms in a forensic context, certain complexities must be understood and addressed. These complexities include: 1) SNPs can generate 0, 1, or arbitrarily many polymorphisms in a polypeptide; and 2) as an object of expression that is modulated by alleles, genes and interactions with the environment, proteins may be present or absent in a given sample. To address these issues, a novel approach was taken to generate the expected protein alleles in a reference sample based on whole genome (or exome) sequence data and assess the significance of the evidence using a haplotype-based semi-continuous likelihood algorithm that leverages whole proteome data. Converting the genomic information into the proteomic information allows for the zero-to-many relationship between SNPs and GVPs to be abstracted away. When viewed as a haplotype, many GVPs that correspond to the same SNP is equivalent to many SNPs in perfect linkage disequilibrium (LD). As long as the likelihood formulation correctly accounts for LD, the correspondence between the SNP and the proteome can be safely neglected. Tests were performed on simulated samples, including single-source and two-person mixtures, and the power of using a classical semi-continuous likelihood versus one that has been adapted to neglect drop-out was compared. Additionally, summary statistics and a rudimentary set of decision guidelines were introduced to help identify mixtures from protein data.

ProSynAR: a reference aware read merger.

MOTIVATION: Read-merging algorithms that look solely at the reads can misalign and mis-merge the reads (especially near repetitive sequences). RESULTS: The C++ program ProSynAR has been written to take the reads' position in the reference into account when performing (and deciding whether to perform) a merge. AVAILABILITY: *Nix users can retrieve the source from GitHub (https://github.com/Benjamin-Crysup/prosynar). Windows binary available at https://github.com/Benjamin-Crysup/prosynar/releases/download/1.0/prosynar.zip. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Evaluation of Promega PowerSeq™ Auto/Y systems prototype on an admixed sample of Rio de Janeiro, Brazil: Population data, sensitivity, stutter and mixture studies.

Forensic DNA typing typically relies on the length-based (LB) separation of PCR products containing short tandem repeat loci (STRs). Massively parallel sequencing (MPS) elucidates an additional level of STR motif and flanking region variation. Also, MPS enables simultaneous analysis of different marker-types - autosomal STRs, SNPs for lineage and identification purposes, reducing both the amount of sample used and the turn-around-time of analysis. Therefore, MPS methodologies are being considered as an additional tool in forensic genetic casework. The PowerSeq™ Auto/Y System (Promega Corp), a multiplex forensic kit for MPS, enables analysis of the 22 autosomal STR markers (plus Amelogenin) from the PowerPlex® Fusion 6C kit and 23 Y-STR markers from the PowerPlex® Y23 kit. Population data were generated from 140 individuals from an admixed sample from Rio de Janeiro, Brazil. All samples were processed according to the manufacturers' recommended protocols. Raw data (FastQ) were generated for each indexed sample and analyzed using STRait Razor v2s and PowerSeqv2.config file. The subsequent population data showed the largest increase in expected heterozygosity (23%), from LB to sequence-based (SB) analyses at the D5S818 locus. Unreported allele was found at the D21S11 locus. The random match probability across all loci decreased from 5.9 × 10-28 to 7.6 × 10-33. Sensitivity studies using 1, 0.25, 0.062 and 0.016 ng of DNA input were analyzed in triplicate. Full Y-STR profiles were detected in all samples, and no autosomal allele drop-out was observed with 62 pg of input DNA. For mixture studies, 1 ng of genomic DNA from a male and female sample at 1:1, 1:4, 1:9, 1:19 and 1:49 proportions were analyzed in triplicate. Clearly resolvable alleles (i.e., no stacking or shared alleles) were obtained at a 1:19 male to female contributor ratio. The minus one stutter (-1) increased with the longest uninterrupted stretch (LUS) allele size reads and according to simple or compound/complex repeats. The haplotype-specific stutter rates add more information for mixed samples interpretation. These data support the use of the PowerSeqTM Auto/Y systems prototype kit (22 autosomal STR loci, 23 Y-STR loci and Amelogenin) for forensic genetics applications.

Graph Algorithms for Mixture Interpretation.

The scale of genetic methods are presently being expanded: forensic genetic assays previously were limited to tens of loci, but now technologies allow for a transition to forensic genomic approaches that assess thousands to millions of loci. However, there are subtle distinctions between genetic assays and their genomic counterparts (especially in the context of forensics). For instance, forensic genetic approaches tend to describe a locus as a haplotype, be it a microhaplotype or a short tandem repeat with its accompanying flanking information. In contrast, genomic assays tend to provide not haplotypes but sequence variants or differences, variants which in turn describe how the alleles apparently differ from the reference sequence. By the given construction, mitochondrial genetic assays can be thought of as genomic as they often describe genetic differences in a similar way. The mitochondrial genetics literature makes clear that sequence differences, unlike the haplotypes they encode, are not comparable to each other. Different alignment algorithms and different variant calling conventions may cause the same haplotype to be encoded in multiple ways. This ambiguity can affect evidence and reference profile comparisons as well as how "match" statistics are computed. In this study, a graph algorithm is described (and implemented in the MMDIT (Mitochondrial Mixture Database and Interpretation Tool) R package) that permits the assessment of forensic match statistics on mitochondrial DNA mixtures in a way that is invariant to both the variant calling conventions followed and the alignment parameters considered. The algorithm described, given a few modest constraints, can be used to compute the "random man not excluded" statistic or the likelihood ratio. The performance of the approach is assessed in in silico mitochondrial DNA mixtures.

ProDerAl: reference position dependent alignment.

MOTIVATION: Current read-mapping software uses a singular specification of alignment parameters with respect to the reference. In the presence of varying reference structures (such as the repetitive regions of the human genome), alignments can be improved if those parameters are allowed vary. RESULTS: To that end, the C++ program ProDerAl was written to refine previously generated alignments using varying parameters for these problematic regions. Synthetic benchmarks show that this realignment can result in an order of magnitude fewer misaligned bases. AVAILABILITY AND IMPLEMENTATION: *Nix users can retrieve the source from GitHub (https://github.com/Benjamin-Crysup/proderal.git). Windows binary available at https://github.com/Benjamin-Crysup/proderal/releases/download/v1.1/proderal.zip. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Reducing noise and stutter in short tandem repeat loci with unique molecular identifiers.

Unique molecular identifiers (UMIs) are a promising approach to contend with errors generated during PCR and massively parallel sequencing (MPS). With UMI technology, random molecular barcodes are ligated to template DNA molecules prior to PCR, allowing PCR and sequencing error to be tracked and corrected bioinformatically. UMIs have the potential to be particularly informative for the interpretation of short tandem repeats (STRs). Traditional MPS approaches may simply lead to the observation of alleles that are consistent with the hypotheses of stutter, while with UMIs stutter products bioinformatically may be re-associated with their parental alleles and subsequently removed. Herein, a bioinformatics pipeline named strumi is described that is designed for the analysis of STRs that are tagged with UMIs. Unlike other tools, strumi is an alignment-free machine learning driven algorithm that clusters individual MPS reads into UMI families, infers consensus super-reads that represent each family and provides an estimate the resulting haplotype's accuracy. Super-reads, in turn, approximate independent measurements not of the PCR products, but of the original template molecules, both in terms of quantity and sequence identity. Provisional assessments show that naïve threshold-based approaches generate super-reads that are accurate (∼97 % haplotype accuracy, compared to ∼78 % when UMIs are not used), and the application of a more nuanced machine learning approach increases the accuracy to ∼99.5 % depending on the level of certainty desired. With these features, UMIs may greatly simplify probabilistic genotyping systems and reduce uncertainty. However, the ability to interpret alleles at trace levels also permits the interpretation, characterization and quantification of contamination as well as somatic variation (including somatic stutter), which may present newfound challenges.

What Happens When Threading is Suppressed in Blends of Ring and Linear Polymers?

Self-diffusivity of a large tracer ring polymer, D r , immersed in a matrix of linear polymers with N l monomers each shows unusual length dependence. D r initially increases, and then decreases with increasing N l . To understand the relationship between the nonmonotonic variation in D r and threading by matrix chains, we perform equilibrium Monte Carlo simulations of ring-linear blends in which the uncrossability of ring and linear polymer contours is switched on (non-crossing), or artificially turned off (crossing). The D r ≈ 6 . 2 × 10 - 7 N l 2 / 3 obtained from the crossing simulations, provides an upper bound for the D r obtained for the regular, non-crossing simulations. The center-of-mass mean-squared displacement ( g 3 ( t ) ) curves for the crossing simulations are consistent with the Rouse model; we find g 3 ( t ) = 6 D r t . Analysis of the polymer structure indicates that the smaller matrix chains are able to infiltrate the space occupied by the ring probe more effectively, which is dynamically manifested as a larger frictional drag per ring monomer.