Table 1.
A summary of research goals and analyses.
Table 2.
K-means selected templates.
Fig 1.
Comparison of RMSEs for the mouse sample.
RMSEs calculated between estimated landmarks and the “gold standard”. Each box represents the RMSEs of 54 mouse specimens calculated by a specific analysis. “ALPACA”: Mouse ALPACA landmarks are estimated using the synthetic mouse template used in the original ALPACA paper. Other columns are ALPACA-derived estimates using each specified template. To see RMSEs expressed as percentage of specimen centroid sizes based on “Gold Standard” manual landmarks, please see S2 Fig.
Fig 2.
Comparison of individual landmark errors for the mouse sample.
Errors represented by Euclidean distances (for errors expressed in percentage of centroid sizes, see S3 Fig). Each column represents errors between MALPACA-derived estimates of all 54 mouse specimens for one landmark and their corresponding Gold Standard (GS) landmarks. Red dots: each represents the median of errors between the estimates of the synthetic template ALPACA of all 54 mouse specimens for one landmark and their corresponding GS landmarks.
Fig 3.
Comparison of the pairwise distances derived from the mouse estimated and GS landmarks.
(A) MALPACA vs. GS. (B) ALPACA vs. GS based on the synthetic template. “r” = correlation coefficient. The plot is based on distances between all pairs of specimens within the mouse sample derived from separate GPA of estimated and GS landmarks.
Fig 4.
Correlations in the first six Principal Components (PCs) between estimated and GS landmarks.
(A) MALPCA vs. GS. (B) ALPACA (the synthetic template) vs. GS. PC scores are derived from separate GPAs of estimated and manual landmarks. Each grid represents the correlation coefficient between a PC from an automated landmarking analysis and a PC from the GS, depending on its row and column names.
Fig 5.
Comparison of RMSEs for the ape sample.
RMSEs calculated between estimated landmarks and the GS landmarks. Each column represents the RMSEs of 46 ape specimens calculated by a specific analysis. “Ape Intraobserver” refers to the RMSEs between two manual landmark datasets of the ape sample. See S4 Fig for RMSEs as percentage of centroid sizes. See S2 Table for the template used for each ALPACA based on a K- means selected template.
Fig 6.
Ape MALPACA-Gold standard correlations in shape variables.
(A) Estimated against GS pairwise Procrustes distances. (B) Correlations in the scores of the first six PCs (principal components) between estimated and GS landmarks. Each grid represents the correlation coefficient between a PC from an automated landmarking analysis and a PC from the GS, depending on its row and column names. Procrustes distances and PC scores are derived from separate GPAs of estimated and manual landmarks.
Fig 7.
Individual landmark errors of the mouse sample comparing to intraobserver errors.
Similar to Fig 2, columns represent errors between MALPACA and the GS for each landmark. Black dots: mean errors of the mouse MALPACA. Green dots: mean intraobserver errors computed by Percival et al. [9]. The labels on the x-axis are landmarks in this study that overlap with those from Percival et al. [9].
Fig 8.
Individual landmark errors of the ape sample.
Errors represented by Euclidean distances (for errors expressed in percentage of centroid sizes, see S3 Fig). Each column represents errors between MALPACA-derived estimates of all 46 ape specimens for the one landmark and their corresponding Gold Standard (GS) landmarks. Each green dot represents the median of the intraobserver manual errors of all 46 ape specimens for one landmark between two manual landmark sets.
Fig 9.
Permutation analyses for evaluating the performance of the K-means template selection method.
(A) The boxplot shows the mean RMSEs of the 100 MALPACA runs from the mouse permutation analysis. The red horizontal line represents the mean RMSE from the mouse K-means based MALPACA. (B) The boxplot shows the mean RMSEs of the 50 MALPACA runs from the ape permutation analysis. The red horizontal line represents the mean RMSE from the ape K-means based MALPACA.
Fig 10.
RMSEs comparison boxplots for ape refined MALPACAs.
(A) The boxplot for comparing the performance of the ape MALPACA and the new median estimates achieved after removing outlier estimates demonstrated in S8 Fig. Comparison is based on calculating RMSEs between estimates and the GS. Two-sided Welch t-test shows that the RMSEs yielded from these two analyses are not significantly different with a p-value 0.5751. (B) RMSEs between the original ape MALPACA and “species-specific” MALPACA for the ape sample. Species-specific MALPACA refers to performing a MALPACA for one species only using templates of that species. RMSEs are calculated between estimates and the GS.