Nick Evershed, Guardian Australia's data journalist, will lead the day looking at how to mine datasets for trends, how to spot and assemble stories, and how to access free tools to show your data.
Reassuringly, we found that there were few substantive differences in our results when considering different methods and datasets; moreover, trends tended to be clearly preserved with the main differences appearing to stem from diminished statistical power when using the smaller data sets.
Figure 5 shows that for majority of real-life datasets, the trend regarding the aforementioned three characteristics is similar to the trend for toy datasets, as indicated by the accompanying gray trend-lines.
As a result, we obtained the best prediction among the regression methods cited above for each term to be used for building the datasets for trend analysis.
Although only one example is provided for both datasets, this trend is clear when considering the sensitivity and specificity values in the function of the similarity threshold, as described above.
Nevertheless, in a number of datasets this trend changes, leading the accuracy level to low values.
In practice, the authors have taken the habit of running their methods on two or three datasets, verifying trend agreement.
Within each dataset, the trend for increasing imputation accuracy with increasing levels of genotyping of immediate ancestors (i.e. parents and grandparents) that was observed for mapped SNP, was also observed for unmapped SNP.
However, in our dataset, this trend is strongly influenced by the single, poorly performing sample, thus it would be recommendable for future research to focus on a more detailed investigation of the robustness of this trend in a larger number of older samples.
Once again, in Table 5 and Fig. 2, we only show the results for specific datasets, but the trends are similar to other datasets as well.
