Run PERMANOVA directly on Wasserstein distance

[ArtPoon]

From reviewer:

A much simpler and standard way to do this is to just put the technical covariates directly into a sequential (Type I) PERMANOVA model of the original, unprojected Wasserstein distance matrix:
adonis2(wdist ~ log(ncod) + log(treelen) + family + exposed * enveloped, data=mdat, by='terms')
This way, the model attributes variance to log(ncod) and log(treelen) first, controlling for them before testing the biological factors, all while using the full distance matrix.

When I run this sequential model on the 205 valid proteins, log(ncod) explains 58.3% of the variance, log(treelen) explains 2.8%, and family explains 12.9%. The biological factors explain a very small slice: exposure explains 0.36% (marginally significant, P = 0.082) and the interaction term explains 0.56% (statistically significant, P = 0.025).

[ArtPoon]

Script provided by reviewer was able to match 238 out of 238 JSON files to metadata. Need to upload JSON files and correct metadata file to Zenodo or this repo.

1. Sequential PERMANOVA on full unprojected Wasserstein distance matrix:
Permutation test for adonis under reduced model
Terms added sequentially (first to last)
Permutation: free
Number of permutations: 9999

adonis2(formula = wdist ~ log(ncod) + log(treelen) + family + exposed * enveloped, data = mdat, permutations = 9999, by = "terms")
                   Df SumOfSqs      R2        F Pr(>F)    
log(ncod)           1  113.269 0.59878 522.0107 0.0001 ***
log(treelen)        1    4.713 0.02491  21.7202 0.0001 ***
family             14   21.616 0.11427   7.1155 0.0001 ***
exposed             1    0.909 0.00481   4.1910 0.0185 *  
exposed:enveloped   1    1.140 0.00602   5.2525 0.0077 ** 
Residual          219   47.520 0.25121                    
Total             237  189.167 1.00000                    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

[ArtPoon]

2. PERMANOVA on 2D residualized distance matrix (exposed * enveloped):
Permutation test for adonis under reduced model
Terms added sequentially (first to last)
Permutation: free
Number of permutations: 9999

adonis2(formula = dmx ~ exposed * enveloped, data = mdat, permutations = 9999, by = "terms")
                   Df SumOfSqs      R2       F Pr(>F)    
exposed             1    0.005 0.00011  0.0326 0.9600    
enveloped           1    4.972 0.12490 35.6001 0.0001 ***
exposed:enveloped   1    2.152 0.05405 15.4050 0.0002 ***
Residual          234   32.682 0.82094                   
Total             237   39.811 1.00000                   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

[ArtPoon]

3. PERMANOVA on 2D residualized distance matrix (family + exposed + enveloped):
Permutation test for adonis under reduced model
Terms added sequentially (first to last)
Permutation: free
Number of permutations: 9999

adonis2(formula = dmx ~ family + exposed + enveloped, data = mdat, permutations = 9999, by = "terms")
          Df SumOfSqs      R2       F Pr(>F)    
family    14   16.863 0.42359 11.6595 0.0001 ***
exposed    1    0.013 0.00033  0.1274 0.8579    
Residual 222   22.934 0.57608                   
Total    237   39.811 1.00000                   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

4. Family-wise PERMANOVA for Picornaviridae (2D Residualized):
Permutation test for adonis under reduced model
Terms added sequentially (first to last)
Permutation: free
Number of permutations: 9999

adonis2(formula = temp_pico ~ exposed, data = mdat[idx_pico, ], permutations = 9999, by = "terms")
         Df SumOfSqs      R2      F Pr(>F)    
exposed   1   0.9446 0.25493 16.082  2e-04 ***
Residual 47   2.7606 0.74507                  
Total    48   3.7052 1.00000                  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1