<i>In silico</i> protein prediction methodology.

We employ a multi-step process which combines computational and manual filtering to identify putative functional candidates in coral. For a protein family of interest, a small number of human proteins are manually selected, and a larger number are computationally selected. We perform two remote homology database searches using HHblits—the manually selected human proteins against the database of coral proteins, and the coral hits against the human protein database. We consider a coral protein a candidate only if it is a best- bidirectional match—if the human protein it matched with was the highest likelihood hit in both directions of the search. The process to this point can be fully automated. However, further verification of the coral candidates requires domain-specific knowledge and involves an in-depth analysis of the structural viability of the candidate through both traditional (see Figs 3 and 4) and machine learning (Figs 5 and 6) methods.