diff --git a/paper/paper.md b/paper/paper.md
index 87bf40d..7dfa193 100644
--- a/paper/paper.md
+++ b/paper/paper.md
@@ -32,12 +32,12 @@ bibliography: paper.bib
 
 Repetitive elements constitute a substantial fraction of most eukaryotic genomes.
 Still, their actual amount differs strongly between species. For example, the genome of *Saccharomyces cervisiae*
-contains only about 3 % repeats ([@kim_transposable_1998]), *Arabidopsis* harbours 14 % ([@the_arabidopsis_genome_initiative_analysis_2000]),
-human 50 % ([@lander_initial_2001]) and wheat even 90 % ([@clavijo_improved_2017]).
+contains only about 3 % repeats [@kim_transposable_1998], *Arabidopsis* harbours 14 % [@the_arabidopsis_genome_initiative_analysis_2000],
+human 50 % [@lander_initial_2001] and wheat even 90 % [@clavijo_improved_2017].
 
 Annotation and Classification of these elements is a pivotal step in the annotation of each genome.
 Furthermore, tracing their history can give ample insights into the evolution of a genome and thereby,
-of a species. Accordingly, different methods for repeat annotation have been developed ([@smit_repeatmasker_2013], [@benson_tandem_1999], [@gymrek_lobstr:_2012]).
+of a species. Accordingly, different methods for repeat annotation have been developed [@smit_repeatmasker_2013; @benson_tandem_1999; @gymrek_lobstr:_2012].
 Still, typically they rely on an assembled genome sequence – a prerequisite which can lead to erroneous results.
 As repetitive elements are highly similar assembly algorithms will collapse repeat variants into a single
 occurrence or not assemble the repetitive regions at all. Thus, the annotation of repeat regions and thereby the
@@ -45,18 +45,18 @@ characterization of their content and diversity solely based on an assembled gen
 
 To address this challenge, we developed reper, a kmer based method to detect, classify and quantify repeats
 in next generation sequencing (NGS) data without the need of a genome assembly.
-Our pipeline samples reads with high kmer coverage directly from the NGS dataset (kmer counts based on jellyfish [@marcais_fast_2011]). This subset is 
-assembled using the transcriptome assembler Trinity ([@grabherr_full-length_2011]), allowing reper to recover repeat variants at a high resolution.
-To create exemplar sequences of each repeat in the genome, the assembled repeats ar clustered using cd-hit ([@li_cd-hit:_2006],[@fu_cd-hit:_2012]).
-These are further classified based on homology to known repeats using multiple blast ([@camacho_blast]) searches. Since reper was developed with
-a focus on plant data, the default classification libraries are REdat ([@nussbaumer_mips_2012]) for repeats, and refseq ([@oleary_reference_2016]) for chloroplast and mitochondrial
+Our pipeline samples reads with high kmer coverage directly from the NGS dataset. The kmer counts are acquired using jellyfish [@marcais_fast_2011]. This subset is 
+assembled using the transcriptome assembler Trinity [@grabherr_full-length_2011], allowing reper to recover repeat variants at a high resolution.
+To create exemplar sequences of each repeat in the genome, the assembled repeats are clustered using cd-hit [@li_cd-hit:_2006; @fu_cd-hit:_2012].
+These are further classified based on homology to known repeats using multiple blast [@camacho_blast] searches. Since reper was developed with
+a focus on plant data, the default classification libraries are REdat [@nussbaumer_mips_2012] for repeats, and refseq [@oleary_reference_2016] for chloroplast and mitochondrial
 sequences. The reference database, however, can easily be customized to the user's needs. A configuration script for
 the popular, but proprietary database repbase is provided with the package as well.
-Next, the repeat content is quantified on sequence, cluster and class level using read mappings (bowtie2 and samtools, [@langmead_fast_2012] and [@li_sequence_2009]).
+Next, the repeat content is quantified on sequence, cluster and class level based on read mappings using bowtie2 and samtools [@langmead_fast_2012; @li_sequence_2009].
 Finally, the repeat landscape can be analyzed and graphically represented with the R script provided with the pipeline.
 Currently, reper is specifically customized to work with paired-end Illumina data, but support of long-read technologies such as PacBio and Nanopore is in development.
 
-To date, there is only a single software package with a similar functionality to reper, namely dnaPipeTE ([@goubert_novo_2015]).
+To date, there is only a single software package with a similar functionality to reper, namely dnaPipeTE [@goubert_novo_2015].
 Still, it relies on dependencies like RepeatMasker, which has to be installed independently as well as the proprietary repeat database repbase by giri.
 Contrasting, The reper source code is available on [github](https://github.com/nterhoeven/reper) under the MIT license.
 To further ease installation and usage, a Docker container with a complete reper installation is also provided.