Impute pairwise amino acid contact probability/energy coefficients

[mikessh]

There are some missing values for amino acid contact probability coefficients and energies

• Contact probabilities are derived from (Is in contact?) ~ Calpha distance + AA1:AA2 coef GLM with a binomial link, see https://github.com/antigenomics/tcr-pmhc-study/blob/master/model/model_simple.Rmd#amino-acid-preferences-and-calpha-distance and https://github.com/antigenomics/tcr-pmhc-study/blob/master/model/eval/aa_pairwise_contact_coef.txt (note the latter is symmetric)
• Energies are computed as mean GROMACS point energy for amino acids that are in contact, see https://github.com/antigenomics/tcr-pmhc-study/blob/master/model/model_simple.Rmd#contact-energies and https://github.com/antigenomics/tcr-pmhc-study/blob/master/model/eval/aa_pairwise_energy.txt

The idea is to model contact probability coefficient/energy with smth like Kidera factors. This way we can both derive factors that influence amino acid preferences in TCR:antigen contacts and compute missing values from Kidera factors of corresponding amino acids.

Not that this problem can perhaps be solved automatically when more data becomes available