Training Sets

Table 4.2: Training sets: number of sequences (n) and nucleotides (nt) used to calculate hexamer dissimilarity. Data are summarized before (raw) and after (filtered) trimming to remove poly-A and poly-T regions, and screening to remove ribosomal RNA sequences.

TAXON	RAW		FILTERED
	n	nt	n	nt
Chytridiomycota	28	52,737	28	50,198
Zygomycota	345	461,423	335	402,834
TOTAL, FUNGI			363	453,032
Daucus	20	98,071	20	64,907
Glycine	411	773,236	411	533,121
Medicago	123	251,882	115	156,884
TOTAL, PLANTS			546	754,912
Bradyrhizobium	165	948,036	165	943,954
Rhizobium	608	2,339,473	607	2,260,533
Sinorhizobium	336	965,446	335	963,287
TOTAL, RHIZOBACTERIA			1107	4,167,774

To characterize hexamer frequencies in plant hosts and their microbial symbionts, we collected sets of training sequences from GenBank and edited them for quality. Using methods described in Section 2.3.5, we queried GenBank to gather training sets of protein-coding sequences from three taxa: ($A$) fungi (Zygomycetes and Chytridiomycetes), ($B_1$) plants ( Daucus, Medicago, and Glycine spp.), and ($B_2$) rhizobacteria (Rhizobium, Sinorhizobium, and Bradyrhizobium spp.). A further constraint was imposed here, that the molecule isolated for sequencing should be DNA, rather than mRNA, to minimize instances of the hexamers AAAAAA and TTTTTT. In the case of eukaryotic taxa, use of DNA sequences to the exclusion of mRNAs implies that non-coding introns are present in the training sets.

Training sets were subjected to the filtering and screening procedures described above, namely that N-rich regions be excluded, poly-A and poly-T stretches be trimmed to no longer than 8 nt, non-protein coding organellar and ribosomal DNA sequences be excluded, and that the final sequence be at least 100 nt in length. The resulting training sets are available as supplementary material, and are summarized in Table 4.2.