This article is from
Journal of Creation 38(1):93–103, April 2024

Browse our latest digital issue Subscribe

Reassessing human–chimpanzee genetic similarity

by Robert W. Carter

The question of how similar humans are to other species has been debated for centuries. In the early 20^th century, most scientists assumed that proteins were the carrier of genetic information, and so the protein content of humans and, for example, apes was assumed to be highly divergent. The discovery that many proteins were similar among the various species, sometimes even identical, came as a shock to many. When molecular methods were first being developed, there was quite a contention within the scientific community, with most thinking that our closest ‘relative’ was the gorilla (Gorilla gorilla) and some believing it was the orangutan (Pongo pygmaeus). King and Wilson published the first human–chimp DNA hybridization experiments in 1975.¹ Sibley and Ahlquist² followed up with more detailed experiments in 1984.³ They showed quite clearly that human DNA was most similar to, first, chimpanzee (Pan troglodytes), then to gorilla, then to orangutans, but this was not universally accepted; they were still defending their results as late as 1990.⁴ The earliest DNA sequencing data concentrated on specific protein-coding genes, which were found to be highly similar in the two species. It was from these earlier studies that we obtained the ‘98% or 99%’ similarity figures that are so often cited. Yet, the true similarity is less than that, a fact that has been known for quite some time.⁵ The notion of a high similarity between the two species is bolstered by the fact that there are, indeed, large areas of high similarity, specifically in the protein-coding regions. Yet, much of the discussion has centred around these places to the exclusion of other genomic compartments that are much less similar.

Worse, DNA hybridization can only test similarity among sequences that will align. When heating up DNA in solution, the strands will separate at a given melting temperature that varies with GC content. The opacity of melted DNA is significantly less than that of aligned DNA, meaning the process can easily be studied in a spectrophotometer. If the DNA of individuals from two different species is mixed and heated, a non-linear reassociation curve with multiple plateaus will be noted as the solution cools and the strands begin to align. By applying a set of complex formulae, and after chemically removing the highly repetitive DNA, the percent similarity of the two species can be estimated.

When Ahlquist realized that he could not tell us how different two species were, only how similar certain portions of their genomes were, he understood that DNA hybridization and reassociation kinetics were extremely limited.⁶ Yes, humans and chimpanzees obviously share a significant portion of highly similar DNA, and estimates put much of that in the 98% similarity range, but there was a large portion of the two genomes that were necessarily excluded from these analyses.

The discussion changed significantly when the first human and chimpanzee genomes were published in 2001⁷ and 2005,⁸ respectively. Various estimates indicated genomic similarity in the 98% range once again. Yet, the human genome was not complete, and the chimpanzee genome was intentionally built using the human genome as a scaffold. The first human genome had 318 long blocks with nothing but the letter ‘N’ (28,000 Ns per block, on average). These spaces mostly covered highly repetitive stretches of DNA that the sequencing technology of the day was unable to handle. The spaces were added to the genome with the hopes that they could later be filled in when better sequencing technology came along. Currently, the most up-to-date version (GRCh38) from the Human Genome Project (HGP) still contains 151 MB of unaligned sequence (approximately 5% of the genome) that has yet to be incorporated into the chromosomes, and the centromeric sequences are fake. That is, “The centromeric alpha satellite arrays are represented as computationally generated models of alpha satellite monomers to serve as decoys for resequencing analyses.” The short arm of chromosome 21 is plagued with problems, and there is evidence of a genome-wide deletion bias.⁹

This hope for a fully sequenced human genome was not realized for over 20 years after the first draft was published (30+ years since the initiation of the HGP). In the summer of 2023, the Telomere-to-Telomere Project (T2T) finally published the full sequence of the last remaining human chromosome, Y.¹⁰

The original chimpanzee genome included the 318 spacer regions seen in the first human genome plus an additional 295,000 smaller ones (average = 51 Ns, calculations below). This was because many short indels (insertions and deletions) must be added if one is to align the two genomes, and the short chimpanzee sequence reads were lined up on the human genome. Coupled to the fact that the first chimpanzee genome was only lightly sequenced (the average coverage was ~5-fold compared to ~30-fold for the human genome), that first attempt gave us a very poor representation of the chimp genome.

Since then, better chimpanzee genomes have been assembled. The first several updates still suffered from the problem of ‘humanization’, but it was eventually assembled without (direct) reference to the human genome. In 2018, Kronenberg et al. published a curated set of nearly 80,000 high-quality chimpanzee contigs (i.e., contiguous stretches of DNA). They used Pac-Bio long-read sequencing technology to get through many of the problematic sections of the chimpanzee genome. Combined with millions of short reads from shotgun sequencing and the testing of specific letters with old-fashioned Sanger sequencing, they managed to achieve approximately 65-fold coverage.¹¹ These contigs were taken from a chimpanzee named Clint. The Clint_ PTRv2 (AKA panTro6) genome was the assembled version of those contigs. Tomkins used a randomly selected set of 18,000 of these chimpanzee contigs and compared them to the assembled genome in his 2018¹² paper. In the meantime, the human genome has gone through multiple rounds of improvement, culminating with T2T (figure 1).

The BLAST program (Basic Local Alignment and Search Tool) is a mainstay of modern genetics. First developed in the early 1990s for searching protein databases for similar sequences, it was rapidly adopted for use with DNA sequences individually (BLASTn) or in batches (MegaBLAST). It uses a heuristic method to make educated guesses about local areas of alignment and is able to find areas of significant similarity about 50 times faster than other, more comprehensive, search algorithms (e.g., Smith-Waterman).

In 2011, Tomkins used BLAST to query 40,000 raw chimpanzee sequence reads against the human genome.¹³ Excluding the areas that did not line up, he estimated 86–89% similarity. Given that BLAST only identifies regions of best alignment, the true similarity should have been less than that. However, it is unclear if those raw sequences reflected an unbiased sampling of the chimpanzee genome.

Tomkins (2011),¹⁴ Tomkins and Bergman (2012),¹⁵ and Bergman and Tomkins (2012)¹⁶ discussed the art of genome construction and multiple frustrations they had with the way the evolutionary community was approaching the subject. In 2013, Tomkins used BLAST to reassess human–chimpanzee sequence similarity.¹⁷ He reached a figure of about 70%. This, however, was due to a glitch in the software being used, as one skeptic claims to have pointed out to Tomkins.¹⁸ By working with the software developers, however, Tomkins was able to get the problem fixed. He then reproduced his original study, this time using a non-buggy algorithm, and arrived at an estimate of 88%.¹⁹ In 2016, he assessed human–chimp similarity by examining 101 trace read data sets from multiple chimpanzee sequencing projects, ‘blasting’ them against the human genome and arriving at an 85% similarity figure.²⁰ In all this work, he was trying to avoid using the chimpanzee genome, since it was demonstrably ‘humanized’. Raw sequence reads might be affected by selection bias and they might have a higher error rate, but they are closer to the source than the assembled genome.

In his latest paper on the subject, Tomkins used BLAST to search for areas of significant similarity using a selection of Kronenberg et al.’s chimpanzee contigs. It took six months of computing time to complete a search of these contigs in the human genome and two versions of the chimpanzee genome. He placed summary tables on GitHub so that anyone could check his results.²¹ By averaging the percent identity (pident) column, he arrived at human–chimpanzee similarity of 85%.

Around this same time, evolutionary geneticist Richard Buggs came out with an estimate of 84.4% similarity, but this was only published in a blog post.²² Later, Seaman and Buggs (2020) published a revised figure of 96.6% using fully aligned genomes, but only after cutting out the centromeres, telomeres, copy number variations, about 300,000 small indels (accounting for about two million letters in each genome), and an additional percentage of DNA that resisted alignment.²³ This ‘apples-to-apples’ comparison is the most robust performed to date, but since they deliberately excluded the most variable portions of the two genomes, the true similarity is necessarily less than 96.6%. How much less is a matter of active investigation.

Several skeptics of Tomkins’ work have complained that he needed to weight his results before calculating any percent similarity.²⁴ While they are technically correct, they have suggested an incorrect method of weighting. Specifically, they noted that Tomkins’ results contained both short and long matches. He simply took the average of all the matches and failed to account for the total length. By taking the total number of aligned bases and dividing by the total match lengths, he would have arrived at a figure closer to 96%. A better method would be to take the match percentage and (conservatively) apply it to the whole contig (not just the matched area), but this produces a comparable similarity score, or one perhaps a few percentage points lower. Even so, both weighting schemes ignore the significant percentage of the genomes that fail to align in BLAST searches.

Yet, any weighting strategy would be inappropriate if the contigs do not represent a fair sampling of the chimpanzee genome. If the database was skewed toward one sequence class over another,²⁵ no amount of ‘weighting’ will help. Thus, instead of weighting by the length of each match, an estimate of the relative frequency of each sequence class represented by the contigs was called for but not performed.

Another major objection is that Tomkins, prior to 2018, chose to use the ungapped feature of BLAST exclusively. This is faster but produces shorter matching regions. However, his critics have promulgated a surprising misunderstanding among themselves. Worse, their purported results seem to back up this misunderstanding, casting doubt on all their calculations and conclusions. Williamson produced an early example which has since been duplicated and even expanded on by others. In an unpublished manuscript²⁶ and a follow-up video,¹⁸ Williamson showed an alignment of two nearly identical short sequences. The only difference was that one had an ‘A’ in the middle, causing the alignment to be perfect for the first half and completely off in the second half (figure 2). He claimed that this would produce a total alignment score of 46%. By inserting a gap in the shorter sequence, however, the alignment score is increased to 92%. Putting aside the fact that he missed one alignable letter (the red line in figure 2), BLAST would actually report a higher percent similarity for the misaligned sequence pair. The algorithm searches for areas with the best local alignment. Thus, it would report back that it had found an area of 100% match for the first sequence pair and only 92% for the other. BLAST does not generally work with such short sequence pairs, but the illustration still holds.

The assertion is that, by disallowing gaps in the search protocol, Tomkins was biasing his results downward. However, there are other reasons why his results are biased downward, and the objection shows a complete misunderstanding of how BLAST works. First, short sequences like this are disqualified. If the matching sequences do not score above some preset minimum (-culling is set to 44 by default), a null result is returned. Second, the -word_size parameter sets the initial minimum match length (default = 11). Once a matching ‘word’ is found, the area is extended to the left and to the right. Each matching letter found increases the score by a set amount (-reward = 2 by default), while each mismatch decreases the score (-penalty = 3 by default). Thus, the bitscore for the match starts out with a value of 22 (word size of 11 × 2 points per matching letter pair) and increases as the alignment is extended. When the score drops to a set amount (-xdrop_ungapped = 20 by default) from any local maximum, the algorithm stops searching, rolls back to the area with the highest score, and reports back that area of alignment only. Thus, it is expected that ‘ungapped’ BLAST searches should produce slightly higher similarity scores than ‘gapped’ searches (figure 3), contrary to Tomkins’ detractors.

An example of scoring in a BLAST search can be seen in figure 3. Here, two identical 1,000-nucleotide sequences were created. An extra letter was then inserted after position 500 in the second string, which threw off the second half of the otherwise perfect alignment. Using -ungapped, BLAST would calculate a maximum score of 1,000 (blue line) and report back a 100% percent identity value for the two strings over a match length of 500. After a single gap is inserted in the shorter string, BLAST would calculate a maximum score of 1,995 (red line) and report back a 99.9% percent identity for the two strings over a match length of 1,000. The descending blue line represents the mistaken notion that BLAST will iterate across the entire query string, in which case it would (falsely) report a bitscore below 200 and a percent identity of 62.3%. However, due to the -xdrop parameter, BLAST will stop searching when the score drops to less than 20 below some local maximum. In this case, the algorithm stops when the score reaches 980, rolls back to the place with the highest score, and reports that it found a 100% match over the first 500 letters. The second half of the string is not tested at all.

BLAST is not intuitive. It takes a brute force approach for finding matches. It will often locate a high-scoring match on the wrong chromosome, and gapped vs. ungapped searches will often hit on very different areas of the genome (see Results). And since the bitscore can rise, even when traversing a ‘gappy’ area with relatively poor alignment, searches that allow for gaps will also often return hits with a lower percent identity than searches that don’t allow for gaps. For these reasons, one must be very careful when trying to estimate total sequence similarity using this program. Worse, BLAST cannot find a sequence match in areas masked-out by the letter ‘N’. Thus, when using a database of sequences that are not incorporated into a genome (e.g., many of the 18,000 contigs used by Tomkins in his 2018 paper had yet to be added to the human genome), BLAST will fail to identify the real matching sequence and will settle on the next-best region, driving down the overall percent similarity. Thus, top-level genomes (which contain only the canonical chromosomes) and full genomes (which also contain unassembled accessory sequences) will not yield the same answers. This is something that has been missed by Tomkins’ detractors. Most of their efforts have focused on top-level genomes, while he made certain to include all the sequence data available.

Methods

The sequences in the contig database used by Tomkins (2018) were obtained from the European Nucleotide Archive²⁷, according to the list he provided. Multiple versions of the chimpanzee and human genomes were obtained (table 1), including the original chimpanzee genome (panTro1, hereafter PT1),²⁸ the Clint_PTRv2 genome (AKA panTro6, hereafter PT2),²⁹ and the most recent version of the chimpanzee genome (panTro3.1.1, hereafter PT3).³⁰ Top-level Genbank version chromosomes were downloaded individually. An additional bulk data download produced an additional 4,300 and 1,446 unaligned sequences for PT2 and PT3, respectively. An early human genome (NCBI34/ hg16, hereafter H16),³¹ a similar version to the one Tomkins used (GRCh37.71, hereafter H37),³² a more recent human genome (GRCh38.p13, hereafter H38),³³ and the Telomere-to-Telomere human genome (hereafter T2T)³⁴ were also obtained. After unzipping, if necessary, chromosome data were concatenated into single FASTA files. Two BLAST databases were created for each genome (one for the chromosomes and one for the unassembled sequences) using the command line.³⁵

The number of N blocks and the total number of Ns were counted for each genome. Using a custom Python script, maps were created for each genome that showed the chromosome lengths and the locations and lengths of all N blocks (figure 1).

Thousands of BLAST searches were performed using a series of custom Python programs. These required the submission of a query sequence, identification of the target database, and the setting of various input parameters (table 2). There are other options available, but not all were tested. Of particular importance was the difference between searches that allowed or disallowed gaps. A ‘gapped’ search is the default, but sending the command -ungapped turns it off. Gapped searches were noticeably slower. The -output_fmt string was set to “10 qid qlen sseqid sstart send pident nident length mismatch gapopen gaps evalue bitscore” (table 3), where ‘10’ just specifies a comma-separated string. The query id, starting location, and length were specified in the BLAST report file name.

In many cases, both ungapped and gapped BLAST searches were performed and compared side-by-side. First, to assess the results of Tomkins (2018), a selection of 150 of the smaller contigs were blasted against PT1, PT2, PT3, H16, H37, H38, and T2T. Several of the most highly repetitive contigs were removed to speed up the analysis (e.g., the time to search varied from a few seconds to a few hours, depending on the repetitiveness of the query). This left 124 contigs and a runtime of approximately 9 hours per genome compared. Second, a 10,000-bp snippet of the longest chimpanzee contig was blasted against PT3 online. This localized it to chimpanzee chromosome 3, so an additional BLAST database was created for this chromosome only. The longest contig (in its entirety) was broken up into pieces 100-, 300-, 1000-, and 10000-bp long and blasted against PT3 chromosome 3. Third, random subsequences of various lengths were chosen from each genome and blasted against other genomes and the parent genome, using a variety of parameter settings. Fourth, the first 500,000 nucleotides of T2T chromosome 22 were broken up into 300-bp and 1,000-bp bins and blasted against PT3 using gapped and ungapped searches.

The Shapiro–Wilk test for normality was applied to the lengths of the contigs, the lengths of the matches in H37, and the pident values for H37 reported by Tomkins (2018), using a Python plugin. A Mann–Whitney U test was used to test for similarity in the normalized pident histograms of the 18,000 contigs and the 124 shorter contigs.

Results

Each of the three tests for normality in Tomkins’ 2018 ‘homo’ data table returned a probability of 0.0. Even though one contig was 2.7 million bases long, the contig lengths were highly skewed toward shorter lengths, with a mode of 1,004 base pairs. The match lengths were equally skewed. The longest match was only 342,000 nucleotides (in a query of nearly two million bp). Matches averaged 62.3% (± 0.31 SD) of the query length, with no clear relationship between query length and match length. The fraction of the query sequence included in the match, however, was highly contingent upon the length of the query and whether or not the search was ungapped or gapped (figure 4). There were zero Ns both among the chimpanzee contigs and within the T2T human genome. The other genomes did not contain any small, sporadic N blocks, as seen in the original chimpanzee genome (table 1). Genome maps are shown in figure 1.

When examining the raw data from Tomkins 2018, the similarity scores seem to come in equally spaced waves, perhaps indicating algorithmic artifacts (figure 5). panTro4 and panTro5 were versions of the chimpanzee genome that were and were not, respectively, assembled using the human genome as a guide. H37 and panTro4 have peaks in highly similar places. Tomkins reported an average pident score for panTro5 of 100%. This could not be replicated either.

The pident scores were not skewed; when plotted as a histogram, they were fully bimodal (figure 4, H37). Tomkins took the average of these values and reported a human–chimpanzee similarity of 84% without accounting for the strange data distribution or the expected genomic frequency of the respective sequence classes within the two main peaks. There were very few values near the ‘average’.

The subset of 124 random small contigs had a highly similar pident distribution to the full collection of 18,000 (figure 4). After normalization, a Mann–Whitney U test performed on the two distributions reported a p-value < 0.00001, meaning the two distributions are essentially identical. Thus, this can be considered a ‘fair sampling’ of the parent distribution and any analyses performed with the subsample should be applicable to the larger set. However, the original numbers could not be validated. When blasted against H37 (the same or similar version of the human genome Tomkins used), most contigs attained a higher pident (figure 6). The results for H37 and H38 were highly similar (figure 7), so the difference between the results of Tomkins (2018) and this study are not likely due to differences in the genome version used. Neither is it expected that different versions of BLAST would produce highly different results (barring programmatic bugs). This discrepancy remains unresolved, although gapped searches did generate results that were closer to Tomkins’ numbers, and he used gapped searches in that study. In essence, the pidents of all low-scoring matches were found at much higher frequencies, especially when using the -ungapped parameter, which he used in earlier studies. Did his use of the gapped parameter drive down the human–chimpanzee similarity in his 2018 study? Attempts were made to recreate his results using various settings of -dust, -soft_masking, and gapping (with identical parameter settings to his 2018 paper). The other user-defined parameters were not expected to make much of a difference.

 

When plotted against Tomkins’ results, many of the new values were higher than he reported (figures 5, 7, and 8). Tomkins’ 2018 data file also does not identify where the matches were located on the human genome, so this cannot be double checked. Also, the current study was unable to reproduce the bimodal peak seen in his data.

Gapped and ungapped searches for matches to those 124 chimpanzee contigs returned nearly identical results for the individual chimpanzee genomes, but the values for PT2 were generally lower than for PT3 (figure 8). It is assumed that this was due to the greater degree of completion of the PT3 genome. Many values went from the 70% range to a full 100% match as the gaps were filled in. Yet, both genomes contained unassembled sequences. It is assumed that the more complete PT3 genome was created by folding in some of the unassembled sequences found in PT2. Thus, the contigs that matched a gap in PT2 should have been located in the accessory sequence data. The reason for the jump in similarity scores is thus unexplained. The ungapped and gapped searches in the two human genomes, however, were split (figure 9). The two ungapped searches were similar and were generally higher than the results for the gapped searches.

Blasting against the various genomes returned high average pident scores. These scores were even higher after weighting was applied (table 4). However, the average length of the matches was drastically different. For the two chimpanzee genomes, a large fraction of the contig was found, on average, to match a section of the chimpanzee genome, though ungapped searches returned smaller match lengths than did gapped searches. For the human genomes, only about a third of the contig, on average, was matched using ungapped searches and just over half of the average contig was captured with gapped searches. Also, many potential matches failed to reach the -culling_limit, the score that must be reached for BLAST to include it on the list of potential hits (default = 44). Being that the -word_size was set to 11, any initial matches automatically start with a score of 22. Only 11 additional matching letters must be added to the seed word to reach a score of 44 (more if mismatches or gaps are found). Fully 14% of ungapped searches against the human genome failed to find any significant matches. In other words, the sequences represented by those contigs do not even exist in the human genome.

Importantly, when blasting a query against its parent genome, BLAST always returned a 100% match for both gapped and ungapped searches.

Many inconsistencies in genome location were noted among the results reported above, so a systematic study of the first 500,000 nucleotides in the T2T chromosome 22 was undertaken. The results of ungapped and gapped searches of T2T against PT3, with two different bin sizes, were highly consistent, but most of the reported ‘best’ matches were not for chromosome 22 (figure 10). Some of this might be due to the translocation of genomic segments among the chromosome arms (whether due to evolution or design). Some might be due to highly similar stretches of DNA being found in more than one place. Much of it might be due to the presence of long and abundant repeats (e.g., Alu elements) that are scattered about the genome. Without a thorough understanding of how BLAST finds comparable sequences, most of the results are probably inapplicable for studying human–chimp differences. Mapping the matches from this section of T2T chromosome 22 onto the PT3 genome reveals the issue starkly (figures 11 and 12). BLAST located parts of this human genome on multiple chimpanzee chromosomes, including several places where consecutive 1,000-bp sections of the two genomes line up beautifully and other places where consecutive 1,000-bp sections of the human genome (falsely) map to the same location in chimpanzee. Worse, that section of overlapping windows moved to another chromosome when switching to a gapped search. These areas (PT3 chromosomes 9:63,565,920–63,566,109 and 14:3,929,276–3,929,593) have been flagged by RepeatMasker.³⁶ BLAST can filter for repetitive sequences (e.g., by setting -dust = yes or -soft_ masking = true), if the sequences are masked (often by setting certain sections to lower case), but doing so had little effect on the results reported above (data not shown).

 

 

Discussion

The high similarity of human and chimpanzee genomes is uncontestable. The evolutionary community has taken this as demonstrable proof of common ancestry. They have a flexible system, though. The date to our most recent common ancestor can shift (and has by several million years over the past several decades), based on fossil or genetic evidence. Yet, being that God clearly created along hierarchical lines,³⁷ there was nothing stopping Him from creating humans and chimpanzees as similar or as dissimilar as He liked. However, chimpanzees and humans have similar behaviours, similar morphology, similar food preferences, and similar temperature requirements. On first principles, therefore, one would expect them to also be similar to us genetically. The answer to the question is not critical for either side, but it is something that many people want to know.

Tomkins’ low estimates were partially driven by incomplete genomic data, but that was all he had to work with at the time. When repeating his methods on more modern genomes, the percent identity of chimpanzees and humans is clearly higher than 85%. The unnoticed bimodal distribution in his pident values should have indicated a problem, but after taking 18,000 readings he felt confident that he had a reasonable sample size, and thus a reasonable average. He also performed a reasonable control test, blasting the contigs against several versions of the chimpanzee genome. With PT2/panTro6, he attained a 100% match average, which is a little strange. When repeated with a smaller sample size of contigs against PT2, many contigs did display a 100% match identity, but certainly not all (table 4, figure 7). The current study also failed to find the peak in pident values he reported that were in the high 60% range when comparing the chimpanzee contigs to human. This was true even when examining the same human genome he used (H37) and a reasonable sampling of his contig database. To date, these discrepancies are unexplained.

His detractors have focused on his lack of weighting and his use of ungapped BLAST searches, but the former is being applied incorrectly, and the latter biases the similarity upward. Gapped searches produce worse matches, as clearly demonstrated here. What is needed is a more comprehensive DNA alignment system. Several have been developed, including Mummer 4,³⁸ LASTZ,³⁹ and Fluent DNA,²³ but these all suffer from assumptions, free parameters, and user input requirements (e.g., gap opening penalties, sensitivity thresholds, and scoring matrixes). They do not just magically pop out the perfect alignment. There is still much ‘art’ to the science of genomic comparison.

LASTZ is probably the most common method used today. Multiple examples of full-genome LASTZ alignments can be found online. The data suggest that long stretches of DNA are shared by humans and chimpanzees (figure 13).⁴⁰ Fluent DNA will only compare genomes that have previously been aligned with other software, but the output data are useful. In their description paper, Seaman and Buggs (2020) presented multiple views and statistics that help us to better assess human–chimpanzee similarity. The oft-cited number from their paper is ‘96.66%’, but that only comes after excluding the centromeres (6.2% of the human genome⁴¹), telomeres (10–15 kb each), copy number variations, masked regions of the input genomes (they compared H38 and PT2, which contained 158 MB and 28 MB of masked regions, respectively), unalignable sequence areas, and all indels (over 2.1 million gaps must be added to each side to align the two genomes). The ignored fraction of the genome totals many millions of bases. The alignment length is only 95.57% of the total reference genome length (H38), so they started off with a substantial difference before the similarity statistic was calculated. Fully 98.65% of the aligned region is identical between humans and chimpanzees. Multiplying these values, approximately 94.27% of the two genomes are identical, and even that needs to be taken with a grain of salt.

BLAST is an inappropriate software platform for making genome-to-genome comparisons, for several reasons. First, it only identifies local matching areas within any given query string, sometimes dropping a significant proportion of the query from the analysis. Second, searches using consecutive strings from the query chromosome do not necessarily locate consecutively on the target chromosome and searches involving highly repetitive sequences will often overlap on the target chromosome. Third, due to the large number of indels that must be added to any multi-species alignment, allowing for gaps in the reported match is necessary, but this tends to lower the average percent similarity. Once a chimpanzee genome is completed in full, it will be possible to create a full-scale alignment between the two genomes. This would need to be manually curated, and differences (i.e., translocations, segmental duplications, gene copy number and placement) would need to be carefully mapped. Additionally, interspecies differences would need to be catalogued. At that point, it would be possible to calculate the full genomic difference between humans and chimpanzees. The value will probably be closer to 95% than to 85%, but as of now there remains a large degree of statistical uncertainty.

Related Media

Four problems with the 99% similar chimp human DNA figure