Protein complex prediction from cores/attachments model
CREATED: 200909250333
State of the art
Markov clustering state of the art 30-50% precision/recall on DIP dataset Gavin et al proposed idea of cores and attachment
Study of benchmark proteins
experimental study: density of complex (0.5633) > average density in interaction graph (0.022) design definition of core and attachment
Algorithm
assign a weight to the edges of PPI (Liu2008, GIW) do MCL clustering filter clusters without cores merge clusters rank them
Evaluation
Jaccard score > 0.5 with benchmark proteins GO term is informative if it has $\ge 30$ proteins and descendant terms have $\lt 30$ proteins (Zhou2002, PNAS) co-locatization score using informative cellular component terms computed as (max overlap over all terms) / (average overlap), from 0 to 1