Identification of bona fide benchmarks to evaluate prediction accuracy in co-fractionation mass spectrometry-based protein complex discovery in plant species. (A) Human and rice proteins are assigned into OGs using the InParanoid algorithm [13]. Rice orthocomplexes are built from CORUM complexes based on the OGs. (B) SEC profiles in CFMS analysis are used in the identification of benchmark complexes. (C) SOM is used to cluster protein profiles generated in B. Each code/group (outer circle) contains a group profile (peak profile), which represents the profiles of all protein members (dots). (D) Subcomplex predictions in SEC datasets are evaluated via a statistical bootstrap P-value calculation. Among experimentally detected subunits (upper panel) in a rice orthocomplex, subunits with similar SEC profiles are clustered in a subcomplex (dotted line). Profile similarity scores are calculated between all possible pairs of subunits in the subcomplex and then are averaged to get the mean dissimilarity of the subcomplex. Simultaneously, an equal number of proteins observed in the orthocomplex are sampled from randomly generated plant orthocomplex (lower panel). The random mean is calculated as mean dissimilarity for pairs of proteins in the random subcomplex. The P-value for each subcomplex is computed as the fraction of times the observed mean is larger than the random mean. (E) Predicted benchmarks can evaluate protein complex prediction results by CFMS performed with any type of biochemical separations.