Table 1.

Accuracy of host prediction based on distance (d) between tetranucleotide frequencies of viral and microbial genomes

DOI: http://dx.doi.org/10.7554/eLife.08490.021

	Predicted	Host order		Host family		Host genus
		Correct	Ratio (%)	Correct	Ratio (%)	Correct	Ratio (%)
All reference sequences
d < 4 × 10−04	98	97	98.98	97	98.98	97	98.98
4 × 10−04 ≤ d < 1 × 10−03	10,173	9361	92.02	8971	88.18	5261	51.72
1 × 10−03 ≤ d	2508	1872	74.64	1757	70.06	917	36.56
Host species excluded
d < 4 × 10−04	21	20	95.24	20	95.24	20	95.24
4 × 10−04 ≤ d < 1 × 10−03	10,003	9067	90.64	8372	83.69	2992	29.91
1 × 10−03 ≤ d	2755	1981	71.91	1840	66.79	818	29.69
Host genus excluded
d < 4 × 10−04	1	0	0.00	0	0.00	0	0.00
4 × 10−04 ≤ d < 1 × 10−03	9085	7303	80.39	6181	68.04	0	0.00
1 × 10−03 ≤ d	3693	1768	47.87	1388	37.58	0	0.00

For each viral genome, the order, family, and genus of its host were predicted from the taxonomy of the closest microbial genome (based on the mean absolute difference between tetranucleotide frequency vectors) and compared to the order, family, and genus of the actual host (i.e., the taxonomy of the genome with which the virus was identified). These predictions were computed with (i) all microbial genomes, (ii) excluding specifically all genomes from the host species, and (iii) excluding all genomes from the host genus. Cases with over 75% of prediction accuracy are highlighted in gray.