A framework for the estimation of the proportion of true discoveries in single nucleotide variant detection studies for human data

Abstract

Any single nucleotide variant detection study could benefit from a fast and cheap method of measuring the quality of variant call list. It is advantageous to be able to see how the call list quality is affected by different variant filtering thresholds and other adjustments to the study parameters. Here we look into a possibility of estimating the proportion of true positives in a single nucleotide variant call list for human data. Using whole-exome and whole-genome gold standard data sets for training, we focus on building a generic model that only relies on information available from any variant caller. We assess and compare the performance of different candidate models based on their practical accuracy. We find that the generic model delivers decent accuracy most of the time. Further, we conclude that its performance could be improved substantially by leveraging the variant quality metrics that are specific to each variant calling tool.

Fig 1. Weighted residuals for model (11, 6) from Table 1.

Fig 2. Hat values for model (11, 6) from Table 1.

Fig 3. Cook’s distances for model (11, 6) from Table 1.

Fig 4. Relationship between Ti/Tv and Het/Hom.

Red, blue and green dots denote Nextera, TrueSeq, and WGS observations, correspondingly. Here one can see an outlying blue point obtained with TrueSeq and Freebayes.

Fig 5. Relationship between AIC and lambda for model (11,6).

AIC values for model (11, 6) are plotted against the parameter lambda used in the variance link function in formula (5). The value of lambda equal to zero corresponds to the log link function in formula (4).

Fig 6. Distribution of the length of 95% PI for PPV.

Because PPV is a proportion, the PI length is measured in %. A box plot of length distribution is provided for each model from Table 1.

Fig 7. Relationship between PPV and Het/Hom.

Fig 8. Relationship between PPV and Ti/Tv.