Editing text

README.md

@@ -162,7 +162,7 @@ Other oddities you might notice are Albacore v0.9.1, which produced a bit more s
 
 This first analysis tackles the most obvious question: how accurate are the basecalled reads? The plot above shows the read identity distribution, with the median (weighted by read length) marked as a horizontal line. Unaligned reads were given an identity of 0% and fall to the bottom of the distribution. Reads with an actual identity below about 65% usually to fail to align and therefore end up at 0%.
 
-Nanonet performed poorly, with a low median and a significant proportion of unaligned reads. Its curiously high peak of about 99% results from its short output sequences discussed above. While a few Nanonet 'reads' did indeed align to the reference with up to 99% identity, these were really small fragments (hundreds of bp) of larger reads.
+Nanonet performed poorly, with a low median and a significant proportion of unaligned reads. Its curiously high peak of about 99% results from its short output sequences discussed above. While a few Nanonet 'reads' did indeed align to the reference with up to 99% identity, these were actually just small fragments (hundreds of bp) of larger reads.
 
 Albacore v0.9.1 and Scrappie raw v1.0.0 had the lowest median identities. While their best reads were comparable to other basecallers' best reads, they produced many more reads below 80%. Excluding those versions, Albacore and Scrappie performed well and were comparable to each other. Scrappie raw v1.1.0 rgr_r94 and Albacore v2.0.2 did best and second-best, respectively. Interestingly, Scrappie produced a significant proportion of unalignable reads in each set, whereas Albacore (excluding v0.9.1) did not – I'm not sure why.
 
@@ -234,7 +234,7 @@ Finally, Nanonet seems a bit dated and should probably be avoided. However, it d
 
 _My_ future work is easy: trying new versions and new basecallers as they are released and adding them to this analysis. Check back occasionally for new data!
 
-The much harder task lies with the basecaller authors: reducing systematic error. As it currently stands, systematic basecalling errors lead to residual errors in assemblies, even after Nanopolish. This makes it hard to recommend an ONT-only approach for many types of genomics where accuracy matters (read more in [our paper on this topic](http://www.biorxiv.org/content/early/2017/07/07/160614)). If systematic error can be eliminated, ONT-only assemblies will approach 100% accuracy, and then ONT will be true Illumina alternative.
+The much harder task lies with the basecaller authors: reducing systematic error. As it currently stands, systematic basecalling errors lead to residual errors in assemblies, even after Nanopolish. This makes it hard to recommend an ONT-only approach for many types of genomics where accuracy matters (read more in [our paper on this topic](http://www.biorxiv.org/content/early/2017/07/07/160614)). If systematic error can be eliminated, ONT-only assemblies will approach 100% accuracy, and then ONT will be a true Illumina alternative.
 
 Did I miss anything important? Can you shed any light on oddities that I couldn't explain? Please let me know through the [issue tracker](https://github.com/rrwick/Basecalling-comparison/issues)!