These days it is very easy to sequence the genome of an unknown organism from short reads, but all subsequent steps - assembly, gene prediction, annotation etc. add quite a bit of noise. Therefore, when someone makes a biological observation from such noisy data, it is important to properly discount it for such uncertainties.
We have been going through a PAML analysis of accelerated evolution and were wondering about how the short read noise would affect the observation of large amount of accelerated evolution. We thoroughly checked all software pipelines to make sure there was no procedural error. However, that did not rule out noise from steps assumed to be accurate by the PAML approach.
To gain full understanding of the mathematics behind the calculation, we went to “Phylogenetic Analysis by Maximum Likelihood (PAML)” website by Ziheng Yang, and also read the 1998 paper “Likelihood Ratio Tests for Detecting Positive Selection and Application to Primate Lysozyme Evolution” by the same author.
Further backtracking on relevant papers got us to other important contributors in the field. One key contributor had been Masatoshi Nei, whose recent book on mutation-driven evolution was covered in our commentary - Darwin Never Proved Natural Selection is the Driving Force of Evolution Because It Isnt. Other major early researchers were W. H. Li and T. Gojobori. Especially, the following 1986 paper - “Simple methods for estimating the numbers of synonymous and nonsynonymous nucleotide substitutions” by Nei and Gojobori - became the basis for many subsequent numerical programs for computing accelerated evolution. Nei and Walter Fitch, who wrote the 1967 Science paper on constructing first phylogenetic tree, also started the respected journal Molecular Biology and Evolution.
The wiki page of Nei led us to one of his early graduate students, who kindly pointed us to his own paper describing exactly what we were worried about. The claims of positive selection are often vastly inflated by errors in sequencing, annotation and alignment as he showed in his 2009 paper. Some of his findings are really alarming (check macaque for example) and even more so, given that his paper covered the era of Sanger sequencing, when genomes were supposedly ‘better’.
Published estimates of the proportion of positively selected genes (PSGs) in human vary over three orders of magnitude. In mammals, estimates of the proportion of PSGs cover an even wider range of values. We used 2,980 orthologous protein-coding genes from human, chimpanzee, macaque, dog, cow, rat, and mouse as well as an established phylogenetic topology to infer the fraction of PSGs in all seven terminal branches. The inferred fraction of PSGs ranged from 0.9% in human through 17.5% in macaque to 23.3% in dog. We found three factors that in?uence the fraction of genes that exhibit telltale signs of positive selection: the quality of the sequence, the degree of misannotation, and ambiguities in the multiple sequence alignment. The inferred fraction of PSGs in sequences that are de?cient in all three criteria of coverage, annotation, and alignment is 7.2 times higher than that in genes with high trace sequencing coverage, known annotation status, and perfect alignment scores. We conclude that some estimates on the prevalence of positive Darwinian selection in the literature may be in?ated and should be treated with caution.
Maybe it is time to have a ‘positive selection-a-thon’ to find out where things are after NGS. For those, who are worried like us about making bad claims about accelerated evolution in papers, Dan Graur suggested to check for the following things -
1. If the sequences that have the lowest sequencing coverage turn out to be the fastest evolving you will know that there is a bias.
2. If the least reliable parts of the alignment tend to evolve the fastest, ditto.
3. Since error tend not to distinguish between synonymous and nonsynonymous substitution, one may find in protein coding genes weird ratios of Ka/Ks in low quality segments, or badly aligned ones.