Development of an Accessible Placental Clock Model

[dw1227]

My supervisor has raised concerns that the top-performing models from the DREAM Placental Clock Challenge may be too complex for broader accessibility within the scientific community. To address this, he has suggested that, in addition to documenting the results of the challenge in the manuscript, I should also develop a placental clock using a more widely recognized and commonly utilized machine learning approach. This model will be built on the entire public dataset, which includes both the training and leaderboard samples from the DREAM challenge.

We expect that the new clock, particularly on the 450k array samples, will outperform the robust placental clock developed by Lee and Horvath since we have increased the sample size, from approximately 1,100 to 1,800 samples.

Tasks:

• [x] Develop a machine learning-based placental clock using familiar methods on the combined public dataset (training + leaderboard).
• [x] Evaluate performance, with the 450k clock expected to exceed the results of the Lee and Horvath placental clock.
• [ ] Package the new model into an accessible R package for public use.

[dw1227]

Now that we have developed the WSU placental clock, Dr. Adi proposed conducting an overrepresentation analysis to identify whether the probes in our model that gain or lose methylation with gestational age are enriched for specific genomic features. This analysis would determine if these probes are disproportionately associated with protein-coding genes, non-coding regions, intergenic areas, promoters, or repetitive/junk DNA regions.