Table 2.
Comparison of the accuracy of inferred transmission networks, summarised over six simulated outbreaks of foot-and-mouth disease in Australia, by transmission network model.
| Algorithm | Accuracya | ||
|---|---|---|---|
| Overall (%) | >50% support (%) | >80% support (%) | |
| Genomic data available for all known infected premises | |||
| Cottam et al.1 | 373/758b (49) | 231/369 (63) | 115/158 (73) |
| Cottam et al. (modified) | 507/758b (67) | 288/369 (78) | 126/158 (80) |
| Gavryushkina et al.5 (SA) | 123/511c (24) | 72/249 (29) | 47/102 (46) |
| Jombart et al.6 (Outbreaker) | 294/764 (38) | 282/665 (42) | 241/490 (49) |
| Lau et al.7 | 559/764 (73) | 519/636 (82) | 406/445 (91) |
| Hall et al.8 (BeastLier) | 118/764 (15) | 66/382 (17) | 31/157 (20) |
| De Maio et al.9 (SCOTTI) | 383/764 (50) | 282/382 (74) | 144/155 (93) |
| Klinkenberg et al.10 (Phybreak) | 363/764 (48) | 222/290 (77) | 112/123 (91) |
| Didelot et al.26 (TransPhylo) | 29/764 (4) | 25/516 (5) | 8/177 (5) |
| Campbell et al.27 (Outbreaker2) | 269/764 (35) | 214/470 (46) | 140/276 (51) |
| Genomic data missing for 50% of known infected premises | |||
| Jombart et al.6 (Outbreaker) | 55/764 (7) | 41/214 (19) | 32/119 (27) |
| Lau et al.7 | 319/764 (42) | 212/356 (60) | 112/154 (73) |
| De Maio et al.9 (SCOTTI) | 93/382d (24) | 73/193 (38) | 37/78 (47) |
| Klinkenberg et al.10 (Phybreak) | 132/764 (17) | 36/50 (72) | 5/7 (71) |
| Campbell et al.27 (Outbreaker2) | 21/764 (3) | 4/50 (8) | 0/20 (0) |
IP = infected premises; SA = Sampled Ancestors. Detailed results per model run provided in Supplementary Materials (S3).
aAccuracy was defined as the proportion of IPs for which the model-predicted most likely source (highest likelihood or most posterior support) was the true source. The denominator for accuracy at > 50% (i.e., consensus) and >80% support includes only those IPs for which the model-predicted most likely source attained that level of likelihood or posterior support. bIn each run, the root was fixed based on best guess. cNot all IPs detected as having sampled ancestors. dSCOTTI only outputs proposed ancestors for those IPs with genomic data available.