Formal Criteria for Gene Families and the Paucity of Large Families in Amborella Compared to All Other Angiosperms

We compare the distribution of gene family sizes in 50 angiosperm genomes. This is based on an automated clustering algorithm for decomposing each genome into gene families.   Chaining is a major problem in the automatic construction of gene families.  Limiting homology-defined families to "cliques" solves this problem but results in another: very small families.  We define measure of "chaininess" to contrast with previous definitions of clique-ness that allow for larger families.  We implement an algorithm for finding "soft cliques with backbones" (SCWiB). This differs from other methods in how it controls chaining, by ensuring clusters satisfy a tolerant edge density criterion that takes into account cluster size.  We compare examples of output from SCWiB and the Markov Cluster Algorithm (MCL), and also compare some curated Arabidopsis thaliana gene families with the results of automatic clustering. The negative correlations between measures of chaininess and cliqueness are confirmed. We apply our method to 50 published angiosperm genomes that have been annotated, and discover that Amborella trichopoda is distinct from all the others in having substantially and systematically smaller proportions of moderate- and large-size gene families.