Notes meeting Daniil 2022-05-12

[richelbilderbeek]

In summary:

• Do a run on the many-SNPs data with only autoencoder functionality for all models #41 Focus on the auto-encoding of the whole genome first. Q: what is the baseline reconstruction of the whole genotype?
  • The primary parameter is the number of neurons in the bottleneck layer
  • [RUNNING] Do a run on the many-SNPs data with for all combinations of autoencoder and phenotypic models #42 by RJCB: when comparing this with GCAE+phenotype, how much does the phenotypic prediction distort the NN?
• From a trained neural network: zoom in on a Region of Interest (ROI), i.e. the spot where a relevant SNP is known to be. Increase the weights for that ROI. Q: is our ROI encoded yes/no?
• From the trained neural network, add the phenotypic prediction

eduPrint_scan_2022-05-12-12-00-31.pdf