To maximize accuracy, the gene structures of B. bassiana were predicted with a combination of different algorithms17 (link)18 (link). The inconsistent ORFs were individually subject to Blast searches against the NCBI curated refseq_protein database and manually inspected. Previously acquired ESTs14 (link) were used to verify and complete the predicted gene models. All predicted gene models were annotated by InterproScan analysis (http://www.ebi.ac.uk/Tools/pfa/iprscan/). The potential secreted proteins were predicted by SignaIP 3.0 (http://www.cbs.dtu.dk/services/SignalP/) and TargetP (http://www.cbs.dtu.dk/services/TargetP/) analysis. Genome repetitive elements were analyzed by Blast against the RepeatMasker library (Open 3.2.9) (http://www.repeatmasker.org/) and with the Tandem Repeat Finder (http://tandem.bu.edu/trf/trf.html). The transposases/retrotransposases were classified by Blastp analysis against the Repbase (http://www.girinst.org/repbase/) plus manual inspections. Putative Beauveria virulence factors were identified by searching against the pathogen-host interaction database (http://www.phi-base.org/about.php) with a cut-off E value of 1e-5, plus additional searches of known virulence genes reported in entomopathogenic fungi. One tail t-tests were conducted to compare the difference of protein family sizes between insect pathogens and other fungi.