Probabilistic expert systems for handling artifacts in complex DNA mixtures.

This paper presents a coherent probabilistic framework for taking account of allelic dropout, stutter bands and silent alleles when interpreting STR DNA profiles from a mixture sample using peak size information arising from a PCR analysis. This information can be exploited for evaluating the evidential strength for a hypothesis that DNA from a particular person is present in the mixture. It extends an earlier Bayesian network approach that ignored such artifacts. We illustrate the use of the extended network on a published casework example.

Estimating mutation rates from paternity casework.

We present a statistical methodology for making inferences about mutation rates from paternity casework. This takes account of a number of sources of potential bias, including hidden mutation, incomplete family triplets, uncertain paternity status and differing maternal and paternal mutation rates, while allowing a wide variety of mutation models. An object-oriented Bayesian network is used to facilitate computation of the likelihood function for the mutation parameters. This can process either full or summary genotypic information, both from complete putative father-mother-child triplets and from defective cases where only the child and one of its parents are observed. We use a dataset from paternity casework to illustrate the effects on inferences about mutation parameters of various types of biases and the mutation model assumed. In particular, we show that there can be relevant information in cases of unconfirmed paternity, and that excluding these, as has generally been done, can lead to biased conclusions.

Identification and separation of DNA mixtures using peak area information.

We introduce a new methodology, based upon probabilistic expert systems, for analysing forensic identification problems involving DNA mixture traces using quantitative peak area information. Peak area is modelled with conditional Gaussian distributions. The expert system can be used for ascertaining whether individuals, whose profiles have been measured, have contributed to the mixture. It can also be used to predict DNA profiles of unknown contributors by separating the mixture into its individual components. The potential of our probabilistic methodology is illustrated on case data examples and compared with alternative approaches. The advantages are that identification and separation issues can be handled in a unified way within a single probabilistic model and the uncertainty associated with the analysis is quantified. Further work, required to bring the methodology to a point where it could be applied to the routine analysis of casework, is discussed.

Probabilistic expert systems for DNA mixture profiling.

We show how probabilistic expert systems can be used to structure and solve complex cases of forensic identification involving DNA traces that might be mixtures of several DNA profiles. In particular, this approach can readily handle cases where the number of contributors to the mixture cannot be regarded as known in advance. The flexible modularity of the networks used also allows us to handle still more complex cases, for example where the finding of a mixed DNA trace is compounded by such features as missing individuals or the possibility of unobserved alleles.

Familial tendency to foetal loss analysed with Bayesian graphical models by Gibbs sampling.

This paper presents several models for investigating whether the HLA allogenotypes DR1/Br, DR3 and DR10 are genetic markers for a predisposition of experiencing unexplained recurrent foetal losses. A total of 199 women from 113 families answered questionnaires concerning their pregnancies and 145 of these women were HLA typed. The analysis of the data is complicated as dependencies between pregnancy outcomes are expected. The main purpose of the paper is to illustrate how such analyses can be performed using Bayesian graphical models and Gibbs sampling. The analyses are made using the programs BUGS and CODA. Markov chain Monte Carlo analyses within a Bayesian framework have become easier with the introduction of these programs. However, experience shows that some caution is required so we recommend making some initial analyses using very simple models and perhaps approximative methods, followed by a model development introducing increasing complexity.

Maternal HLA class II allogenotypes are markers for the predisposition to fetal losses in families of women with unexplained recurrent fetal loss.

The HLA allogenotypes DR1/Br, DR3 and DR10 (entitled risk HLA markers) have been reported as being genetic markers for the predisposition to experience unexplained recurrent fetal losses. The aim of the study was to determine whether the putative risk HLA markers might also be markers for the risk of pregnancy loss in sisters and wives of brothers of women with unexplained recurrent fetal losses. Information concerning pregnancy outcomes among the relatives of 146 consecutive women with unexplained recurrent fetal losses was collected. Ninety-five of the full sisters, 69 of the full brothers and 50 of the wives of the brothers were HLA typed. Sisters who had experienced at least one previous pregnancy loss (affected women) shared more HLA haplotypes with the proband than unaffected sisters, when the proband was positive for the risk markers (P = 0.02). More affected than unaffected sisters and brothers' wives were positive for the risk markers (P < 0.005 and P < 0.03; respectively). The lowest estimate of the odds ratio for experiencing pregnancy loss among sisters and brothers' wives who were positive compared with those negative for the risk markers was 3.5 (95% credible interval = 1.9-5.8). It is concluded that maternal DR1/Br, DR3 and DR10 allogenotypes seem to be genetic markers for the risk of pregnancy loss among relatives of women with unexplained recurrent fetal losses. The pattern of inheritance suggests a polygenic mode of inheritance with alleles linked to the risk HLA markers interacting with non-HLA linked genes expressed on the fetus or the trophoblast.

The reliability of measuring left ventricular ejection fraction by radionuclide cardiography: evaluation by the method of variance components.

A statistical model based on the method of variance components was applied to obtain confidence statements for single and repeat determinations of left ventricular ejection fraction by radionuclide techniques. With this approach variance caused by individual factors in the measurement procedure is estimated to allow calculation of confidence intervals based on single measurements and the detection limits for changes. Six study groups made up of a total of 143 subjects were examined by both multigated equilibrium and first pass imaging. Under favourable conditions (with an updated gamma camera and experienced observer) the 95% confidence interval with a single measurement of left ventricular ejection fraction by equilibrium imaging was +/- 3 ejection fraction units, compared with +/- 6 units with the first pass technique (one ejection fraction unit = 1/100 of the possible values from 0.00 to 1.00). The minimal significant changes (at the 5% level) in measured equilibrium left ventricular ejection fraction at intervals of 15 min, 3 days, 1, 3, and 4 weeks were +/- 4, +/- 4, +/- 5, +/- 5, and +/- 6 units, respectively. The corresponding minimal detectable changes in a subject's "true" left ventricular ejection fraction for the same intervals were +/- 7, +/- 7, +/- 10, +/- 10, and +/- 12 units respectively. With first pass imaging, only average values for the variation at repeat determination could be calculated. The minimal significant change in measured first pass left ventricular ejection fraction was +/- 7 units, and the minimal detectable change in "true" left ventricular ejection fraction was +/- 14 units. Measurements of left ventricular ejection fraction by equilibrium technique were generally more reproducible than first pass determinations because the variability caused by study acquisition, observer analysis, and residual errors was smaller. The method of variance components appears to be well suited to the evaluation of quantitative biological measurements in clinical use. The popularity of established procedures may obscure the lack of basic information about method evaluation.

Reproducibility of determination of right ventricular ejection fraction by radionuclide imaging: assessment by the statistical method of variance components.

Confidence limits for single and repeat measurements of right ventricular ejection fraction (RVEF) were established by means of a model based on the statistical method of variance components. A total of 80 subjects (age 23 to 74 years) were examined by two radionuclide methods 1) gated first-pass (fp) technique performed in a standard 30 degrees right anterior oblique projection, and 2) multigated equilibrium imaging (muga) in an individual left anterior oblique view, applying with both methods separate end-diastolic and end-systolic ventricular regions of interest. Values obtained by fp technique were clearly higher than those measured by the muga approach, and the correlation between them was only fair: RVEFmuga = 0.48 RVEFfp + 0.13; r = 0.73; SEE = 0.08. The 95% confidence limits for a single measurement were with the fp technique: 'true' RVEF = measured RVEF +/- 6 EF-units compared to +/- 16 units with the muga method. At repeat determination within an interval of four weeks, the minimal changes in measured RVEF that were statistically significant at the 5% level were with the fp technique +/- 8 units with the same observer on both occasions and +/- 9 units with different observers. Corresponding figures with the muga method were +/- 16 and +/- 22 units, respectively. The minimal changes in a subject's 'true' RVEF necessary to produce a significant change in measured RVEF were with fp technique +/- 14 units for the same observer and +/- 17 units for different observers, compared to +/- 30 and +/- 41 units with the muga method. In conclusion, the variability with the muga approach was far greater than with the fp technique and the consequent reproducibility so poor as to preclude meaningful measurement of RVEF by the muga method.

Clinical assessment of disease activity in rheumatoid arthritis.

A new experimental design was developed to study the value of clinical parameters of disease activity in patients with rheumatoid arthritis. Ten patients with classical rheumatoid arthritis were examined by five senior doctors in a department of medical rheumatology. In spite of an attempt to make the clinical examination as uniform as possible the inter-observer variation among the doctors was greater than the variation among the patients, for the following parameters: joint pain at rest, joint tenderness and joint swelling. An acceptable inter-observer variation in relation to patient variation was found for 1) a combined registration of joint pain at rest or on movement, 2) duration of morning stiffness, 3) grip strength, 4) subjective well-being as indicated on a visual analogue scale, 5) fingertip--palm distance, and 6) maximum flexion-extension in elbows, wrists and knees. The variation from morning to afternoon and from day to day was negligible. It is concluded that registration of elaborated articular scores is useless in the daily routine in rheumatological departments when different doctors examine the patients.