paper: small edits

paper/paper.md

@@ -51,7 +51,7 @@ The NCBI Taxonomy Database [@schoch_ncbi_2020] provides a hierarchically arrange
 
 The NCBI BLAST+ suite is the most widely used toolset in bioinformatics for performing sequence similarity search [@camacho_blast_2009]. The suite provides a Bash script (`get_species_taxids.sh`) to convert NCBI Taxonomy identifiers (TaxIDs) or text into TaxIDs suitable for filtering sequence searches. While this is a useful utility, it only works with sequences submitted to GenBank or other NCBI-hosted databases, and more importantly, it relies on making API calls via Entrez Direct (EDirect) [@kans_entrez_2024]. EDirect requires an internet connection and it does not scale well when working with large sequence datasets. Other general-purpose tree libraries exist for Python (e.g. ``anytree`` [@anytree] and ``bigtree`` [@bigtree]) and R (e.g. ``ggtree`` [@yu_ggtree_2017]), but they do not support the core features provided by Taxonomy Resolver or focus mainly on tree visualisation. The development of Taxonomy Resolver started in 2020 and aims to provide user-friendly interfaces for working directly with the NCBI Taxonomy hierarchical dataset.
 
-# Implementation
+# Features
 
 Taxonomy Resolver has been developed with simplicity in mind and it can be used both as a standard Python module or as a CLI application. The main tasks performed by Taxonomy Resolver are:
 
@@ -63,9 +63,10 @@ Taxonomy Resolver has been developed with simplicity in mind and it can be used
 * **filtering** a tree based on the inclusion and/or exclusion of certain TaxIDs
 * **writing and loading** tree data structures using Python’s object serialisation
 
-A taxonomy tree is a hierarchical structure that can be seen as a collection of deeply nested containers - nodes connected by edges, following the hierarchy, from the parent node - the root, all the way down to the children nodes - the leaves. An object-oriented programming (OOP) tree implementation based on recursion does not typically scale well for large trees, such as the NCBI Taxonomy, which is composed of >2.6 million nodes. To improve performance, Taxonomy Resolver represents the tree structure following the Nested Set Model, which is a technique developed to represent hierarchical data in relational databases lacking recursion capabilities. This allows for efficient and inexpensive querying of parent-child relationships. The full tree is traversed following the Modified Preorder Tree Traversal (MPTT) strategy [@celko_chapter_2004], in which each node in the tree is visited twice. In a preorder traversal, the root node is visited first, then recursively a preorder traversal of the left sub-tree, followed by a recursive preorder traversal of the right subtree, in order, until every node has been visited. The modified strategy allows capturing the 'left' and 'right' ($lft$ and $rgt$, respectively) boundaries of each subtree, which are stored as two additional attributes. Finding a subtree is as simple as searching for the nodes of interest where $lft > node's\ lft$ and $rgt < node's\ rgt$. Likewise, finding the full path to a node is as simple as searching for the nodes where $lft < node's\ lft$ and $rgt > node's\ rgt$. Traversal attributes, depth and node indexes are captured for each tree node and are stored as a pandas DataFrame [@pandas_2024].
+# Implementation
+
+A taxonomy tree is a hierarchical structure that can be seen as a collection of deeply nested containers - nodes connected by edges, following the hierarchy, from the parent node - the root, all the way down to the children nodes - the leaves. An object-oriented programming (OOP) tree implementation based on recursion does not typically scale well for large trees, such as the NCBI Taxonomy, which is composed of >2.6 million nodes. To improve performance, Taxonomy Resolver represents the tree structure following the Nested Set Model, which is a technique developed to represent hierarchical data in relational databases lacking recursion capabilities. This allows for efficient and inexpensive querying of parent-child relationships. The full tree is traversed following the Modified Preorder Tree Traversal (MPTT) strategy [@celko_chapter_2004], in which each node in the tree is visited twice. In a preorder traversal, the root node is visited first, then recursively a preorder traversal of the left subtree, followed by a recursive preorder traversal of the right subtree, in order, until every node has been visited. The modified strategy allows capturing the 'left' and 'right' ($lft$ and $rgt$, respectively) boundaries of each subtree, which are stored as two additional attributes. Finding a subtree is as simple as searching for the nodes of interest where $lft > node's\ lft$ and $rgt < node's\ rgt$. Likewise, finding the full path to a node is as simple as searching for the nodes where $lft < node's\ lft$ and $rgt > node's\ rgt$. Traversal attributes, depth and node indexes are captured for each tree node and are stored as a pandas DataFrame [@pandas_2024].
 
-# Conclusion
 Taxonomy Resolver has been developed to take advantage of the Nested Set Model tree structure, so it can perform fast validation and create lists of taxa that compose a particular subtree. Inclusion and exclusion lists can also be seamlessly used to produce subset trees with wide applications, particularly for sequence similarity search. Taxonomy Resolver has been in production since 2020 serving thousands of users every month. It provides taxonomy filtering features for NCBI BLAST+ provided by the popular EMBL-EBI Job Dispatcher service, available from https://www.ebi.ac.uk/jdispatcher/sss/ncbiblast [@madeira_2024].
 
 # Acknowledgements