Improved logistic model for combining scores

[damiansm]

Logistic modelling to combine all scores

• should do PCA on all variables to see best ones to include
• NB Can train defaults for non-missense pathogenicity scores as well by setting categorical vals in training set
• NB Tried to use logistic model to combine the human,mouse,fish and PPI scores into a single phenotype score that takes all evidence into account but without success. Problem was the missing data e.g. a gene may not have a mouses or fish model. Tried using average scores for these but made even worse

[damiansm]

Also need to look at retraining for ReMM + pheno combining. CADD alone performs a lot worse on the 414 variants but when combined performance is not much worse than for ReMM. Suggests can be optimised

[damiansm]

Maybe not worth doing if switch to Bayesian method