deletions | additions       diff --git a/section_Network_architecture_begin_figure__.tex b/section_Network_architecture_begin_figure__.tex  index f21d099..9947dd1 100644  --- a/section_Network_architecture_begin_figure__.tex  +++ b/section_Network_architecture_begin_figure__.tex 
...
\centering  \includegraphics[scale=0.5]{figures/mhcflurry-gliffy-network.png}  \caption{Neural network architecture for predicting peptide-MHC affinities from fixed length amino acid sequences}  \end{figure} \section{Predicting affinities for multiple peptide lengths using a 9mer encoding}  Reduction from multiple peptide lengths to a 9mer encoding was done using a scheme inspired by NetMHC\cite{lundegaard2008accurate}. Peptides with only 8 amino acids were extended with the insertion of a special wildcard ``X'' at every position in the sequence. Peptides longer than 9 amino acids were shortened by removing consecutive stretches of residues at every position. These lengthened or shortened samples were included in the training set with a sample weight inversely proportional to the number of samples created from a single measurement.