If your thought when you read someone saying that they’ve overturned 100 years of established science, is “oh wow! This guy’s a genius, they’re all wrong!” And not “what’s this guy missing that everybody else isn’t”, then you really need to address your biases, and how they’re affecting your ability to learn.
As I pointed out to Erika herself, regardless of whether I am an arrogant lunatic or a world-class scientific genius of historic note, how would anyone, no matter how well-credentialed, ever be able to tell the difference given their absolute refusal to even look at the growing body of evidence that natural selection is a very weak force totally incapable of accounting for the chimp-human genetic divergence, much less the origin of the species?
With regards to the inverse of the question concerning what I am missing that everyone else isn’t, Athos and I ran into a very interesting anomaly while we were sorting through the various ape genomes to determine precisely how genetically divergent the seven species are. As it turns out, discovered that the protein-guided CDS alignment method has a circular reasoning problem that’s impossible to ignore once you recognize it:
- You want to measure how conserved proteins are (dN/dS) to prove the strength of purifying natural selection.
- You align them using substitution matrices built from observed conservation patterns
- The aligner penalizes divergent amino acid substitutions while scoring the alignments
- Regions too divergent to score well get gapped out or are left unaligned, so they aren’t counted.
- You measure conservation on what remains
- Hey presto! You find strong conservation! Quelle surprise!
In other words, the method presupposes the conclusion. This is, again, the same way one proves that Kevin Hart can dunk a basketball on a 10-foot basket, by showing that he is one inch taller than Lebron James. The aligner’s scoring function is a model of purifying selection that treats radical amino acid changes as improbable and therefore unlikely to be true matches. So divergent codons are excluded from the alignment before dN/dS is ever computed. Those excluded codons are enriched for nonsynonymous substitutions by definition, because that’s why they scored poorly in the protein alignment.
The whole-genome alignment we utilize doesn’t have this scoring bias. Cactus aligns DNA based on genomic context — synteny, flanking sequence, structural features — without reference to protein function. It’s agnostic about whether a coding region is conserved or divergent. When we project gene coordinates onto that alignment and extract codons, we get ALL the aligned coding positions, not just the ones that are deemed to look conserved at the protein level.
The diagnostic data confirms this. Single-SNV codons, the positions where there’s exactly one nucleotide difference in the codon, so classification is completely unambiguous, show 71.6% nonsynonymous, 28.4% synonymous. dN/dS = 0.71 on those alone. There’s no misalignment artifact that explains a 3:1 nonsynonymous-to-synonymous ratio on unambiguous single-substitution codons. That’s just what the genome looks like.
And so the gene set comparison nails the coffin shut on the alternative explanation. We ran the same test on their smaller subset, which I suspected might be cherry-picked. Except they weren’t. If the protein aligner were simply being more careful and excluding messy regions that our pipeline misaligns, then we’d expect the nonsynonymous fraction to vary considerably across gene subsets. Fast-evolving gene families and duplications would show different ratios than conserved 1:1 orthologs. But they don’t. It’s 73.8% nonsynonymous across every subset we tested. The signal is uniform across the genome, it’s not concentrated in problematic genes.
So what the Consortium’s method actually measured is: “among the coding positions conserved enough to survive protein alignment, how conserved are they?” That’s not a valid measurement, that’s a tautological quantification.
The whole field of evolutionary biology has interpreted dN/dS = 0.23 as evidence that purifying selection eliminates 77% of nonsynonymous mutations before they reach fixation. But the whole-genome measurement of dN/dS = 0.7989 implies that only 20 percent are actually eliminated, which is nearly four times less purifying selection than the mainstream science assumes. The difference is not a matter of gene selection, quality filtering, or computational error. It is a methodological artifact of using a conservation-dependent alignment to measure conservation.
| Gene Set | Genes Measured | Pooled dN/dS | Nonsyn Fraction |
|---|---|---|---|
| A: All GENCODE | 20,400 | 0.7989 | 73.9% |
| B: 1:1 chimp orthologs | 16,995 | 0.7956 | 73.8% |
| B-HC: High-confidence 1:1 | 15,937 | 0.7939 | 73.8% |
| C: Excluded by filter | 3,405 | 0.8237 | 74.4% |
| Published (Consortium 2005) | 13,454 | 0.2300 | ~40% |
These findings have two significant implications.
First, they demonstrate that the canonical evidence for strong purifying selection in coding sequences is methodologically compromised and therefore incorrect. Second, they indicate that even coding sequences are evolving near-neutrally. Purifying selection exists, but is a much weaker mechanism than previously believed, removing roughly one in five nonsynonymous mutations rather than three in four. This is consistent with the drift-dominated picture that emerges from the LTEE data and from the mathematical breeding reality constraints: natural selection exists but operates at the margins, and contrary to what has been believed for the last 160 years, it has never been the dominant force shaping protein evolution between closely related species.
And, just to be clear, none of this genomic analysis has anything to do with MITTENS per se. It merely happens to be empirical evidence that tends to independently confirm the mathematical and empirical impossibility of evolution by natural selection.