add arnaud.

[chloroplast13.git] / annotated.tex

These  last years  the cost  of  sequencing genomes  has been  greatly
reduced,  and thus  more and  more genomes  are  sequenced.  Therefore
automatic annotation tools are required to deal with this continuously
increasing amount of genomical data. Moreover, a reliable and accurate
genome  annotation  process  is  needed  in order  to  provide  strong
indicators for the study of life\cite{Eisen2007}.

Various  annotation   tools  (\emph{i.e.},  cost-effective  sequencing
methods\cite{Bakke2009}) producing genomic  annotations at many levels
of detail  have been designed  by different annotation  centers. Among
the major annotation  centers we can notice NCBI\cite{Sayers01012011},
Dogma       \cite{RDogma},       cpBase      \cite{de2002comparative},
CpGAVAS                   \cite{liu2012cpgavas},                   and
CEGMA\cite{parra2007cegma}. Usually, previous  studies used one out of
three methods  for finding  genes in annoted  genomes using  data from
these  centers: \textit{alignment-based},  \textit{composition based},
or a  combination of both~\cite{parra2007cegma}.   The alignment-based
method  is used  when trying  to predict  a coding  gene (\emph{i.e.}.
genes that produce proteins) by aligning a genomic DNA sequence with a
cDNA  sequence  coding  an homologous  protein  \cite{parra2007cegma}.
This approach is  also used in GeneWise\cite{birney2004genewise}.  The
alternative   method,   the    composition-based   one   (also   known
as  \textit{ab initio})  is based  on  a probabilistic  model of  gene
structure  to  find genes  according  to  the  gene value  probability
(GeneID \cite{parra2000geneid}).  Such  annotated genomic data will be
used to overcome  the limitation of the first  method described in the
previous section.   In fact, the  second method we propose  finds core
genes  from  large  amount  of  chloroplast  genomes  through  genomic
features extraction.

Figure~\ref{Fig1} presents an overview  of the entire method pipeline.
More    precisely,    the   second    method    consists   of    three
stages:   \textit{Genome    annotation},   \textit{Core   extraction},
and    \textit{Features    Visualization}    which   highlights    the
relationships.  To  understand the  whole core extraction  process, we
describe briefly each  stage below. More details will  be given in the
coming subsections.   The method uses as starting  point some sequence
database  chosen  among   the  many  international  databases  storing
nucleotide sequences, like  the GenBank at NBCI \cite{Sayers01012011},
the    \textit{EMBL-Bank}     \cite{apweiler1985swiss}    in    Europe
or   \textit{DDBJ}   \cite{sugawara2008ddbj}   in  Japan.    Different
biological tools can analyze  and annotate genomes by interacting with
these databases to  align and extract sequences to  predict genes. The
database in  our method must be  taken from any  confident data source
that stores annotated and/or unannotated chloroplast genomes.  We have
considered the GenBank-NCBI \cite{Sayers01012011} database as sequence
database:  99~genomes of chloroplasts  were retrieved.   These genomes
lie in  the eleven type  of chloroplast families and  Table \ref{Tab2}
summarizes their distribution in our dataset.\\

\begin{figure}[h]  
  \centering
    \includegraphics[width=0.8\textwidth]{generalView}
\caption{A general overview of the annotation-based approach}\label{Fig1}
\end{figure}

Annotation,  which  is the  first  stage,  is  an important  task  for
extracting gene features. Indeed, to extract good gene feature, a good
annotation tool  is obviously  required. To obtain  relevant annotated
genomes, two annotation  techniques from NCBI and Dogma  are used. The
extraction of gene feature, the  next stage, can be anything like gene
names,  gene  sequences, protein  sequences,  and  so  on. Our  method
considers gene  names, gene counts,  and gene sequence  for extracting
core  genes and  producing  chloroplast evolutionary  tree. The  final
stage   allows  to   visualize  genomes   and/or  gene   evolution  in
chloroplast.    Therefore   we   use  representations   like   tables,
phylogenetic  trees,  graphs,  etc.   to  organize  and  show  genomes
relationships,  and  thus  achieve   the  goal  of  representing  gene
evolution.   In addition,  comparing these  representations  with ones
issued from  another annotation tool dedicated to  large population of
chloroplast genomes  give us biological perspectives to  the nature of
chloroplasts evolution. Notice that  a local database linked with each
pipe stage is  used to store all the  informations produced during the
process.

\input{population_Table}
        
\subsection{Genome annotation techniques}

For  the first  stage, genome  annotation, many  techniques  have been
developed  to annotate chloroplast  genomes.  These  techniques differ
from  each others  in  the number  and  type of  predicted genes  (for
example:  \textit{Transfer  RNA   (tRNA)}  and  \textit{Ribosomal  RNA
(rRNA)}  genes). Two  annotation techniques  from NCBI  and  Dogma are
considered to analyze chloroplast genomes.

\subsubsection{Genome annotation from NCBI} 

The objective  is to generate sets  of genes from each  genome so that
genes are organized  without any duplication.  The input  is a list of
chloroplast genomes  annotated from NCBI. More  precisely, all genomes
are stored as \textit{.fasta} files  which consists in a collection of
protein  coding genes\cite{parra2007cegma,RDogma}  (gene  that produce
proteins) organized in coding sequences.   To be able build the set of
core    genes,     we    need    to     preprocess    these    genomes
using  \textit{BioPython}  package \cite{chapman2000biopython}.   This
step  starts by  converting  each  genome from  FASTA  file format  to
GenVision \cite{geneVision}  format from DNASTAR. Each  genome is thus
converted in  a list of genes,  with gene names and  gene counts. Gene
name duplications can be accumulated during the treatment of a genome.
These  duplications  come   from  gene  fragments  (\emph{e.g.}   gene
fragments treated  with NCBI) and  from chloroplast DNA  sequences. To
ensure that  all the  duplications are removed,  each list of  gene is
translated  into a  set of  genes.  Note that  NCBI genome  annotation
produces genes except \textit{Ribosomal (rRNA)} genes.

\subsubsection{Genome annotation from Dogma}

Dogma stands for \textit{Dual  Organellar GenoMe Annotator}.  It is an
annotation tool  developed at  University of Texas  in 2004  for plant
chloroplast and  animal mitochondrial genomes.  This tool  has its own
database  for translating  a  genome  in all  six  reading frames  and
queries     the     amino     acid     sequence     database     using
BLAST  \cite{altschul1990basic}  (\emph{i.e.}   Blastx)  with  various
parameters.  Protein  coding genes are  identified in an  input genome
using sequence similarity of genes  in Dogma database.  In addition in
comparison   with   NCBI    annotation   tool,   Dogma   can   produce
both \textit{Transfer RNAs (tRNA)} and \textit{Ribosomal RNAs (rRNA)},
verify their start and end  positions. Another difference is also that
there  is   no  gene  duplication   with  Dogma  after   solving  gene
fragmentation. In  fact, genome annotation  with Dogma can be  the key
difference when extracting core genes.

The Dogma  annotation process  is divided into  two tasks.   First, we
manually annotate chloroplast genomes using Dogma web tool. The output
of this step is supposed to  be a collection of coding genes files for
each genome, organized in GeneVision file. The second task is to solve
the  gene   duplication  problem  and   therefore  we  have   use  two
methods. The first method, based  on gene name, translates each genome
into a set  of genes without duplicates. The  second method avoid gene
duplication  through a  defragment  process. In  each iteration,  this
process  starts by taking  a gene  from gene  list, searches  for gene
duplication, if a duplication is found, it looks on the orientation of
the  fragment sequence.   If it  is positive  it appends  directly the
sequence to  gene files.  Otherwise reverse  complement operations are
applied  on the sequence,  which is  then also  append to  gene files.
Finally, a  check for missing start  and stop codons  is performed. At
the  end  of  the  annotation  process,  all  the  genomes  are  fully
annotated,  their   genes  are  defragmented,  and   gene  counts  are
available.

\subsection{Core genes extraction}

The goal of  this stage is to extract maximum core  genes from sets of
genes.  To find core genes, the following methodology is applied.

\subsubsection{Preprocessing}

In order  to extract  core genomes in  a suitable manner,  the genomic
data are preprocessed with two methods: on the one hand a method based
on gene  name and count,  and on  the other hand  a method based  on a
sequence quality control test.

In the first method, we extract  a list of genes from each chloroplast
genome.  Then we store this list of genes in the database under genome
nam and  genes counts can be  extracted by a  specific length command.
The \textit{Intersection  Core Matrix}, described  in next subsection,
is then  computed to  extract the core  genes.  The problem  with this
method can be stated as follows: how can we ensure that the gene which
is  predicted in  core genes  is the  same gene  in leaf  genomes? The
answer  to this problem  is that  if the  sequences of  any gene  in a
genome annotated  from Dogma  and NCBI are  similar with respect  to a
given  threshold,  then   we  do  not  have  any   problem  with  this
method. When the sequences are  not similar we have a problem, because
we cannot decide which sequence belongs to a gene in core genes.

The second method is based on  the underlying idea: we can predict the
the best annotated  genome by merging the annotated  genomes from NCBI
and Dogma according to a quality test on genes names and sequences. To
obtain all  quality genes  of each genome,  we consider  the following
hypothesis: any gene  will appear in the predicted  genome if and only
if the  annotated genes  in NCBI and  Dogma pass a  specific threshold
of  \textit{quality  control test}.   In  fact,  the Needle-man  Wunch
algorithm  is applied  to compare  both  sequences with  respect to  a
threshold. If  the alignment  score is above  the threshold,  then the
gene will be  retained in the predicted genome,  otherwise the gene is
ignored.   Once    the   prediction   of   all    genomes   is   done,
the \textit{Intersection Core Matrix} is computed on these new genomes
to extract core genes, as explained in Algorithm \ref{Alg3:thirdM}.

\begin{algorithm}[H]
\caption{Extract new genome based on gene quality test}
\label{Alg3:thirdM}
\begin{algorithmic} 
\REQUIRE $Gname \leftarrow \text{Genome Name}, Threshold \leftarrow 65$
\ENSURE $geneList \leftarrow \text{Quality genes}$ 
\STATE $dir(NCBI\_Genes) \leftarrow \text{NCBI genes of Gname}$
\STATE $dir(Dogma\_Genes) \leftarrow \text{Dogma genes of Gname}$
\STATE $geneList=\text{empty list}$
\STATE $common=set(dir(NCBI\_Genes)) \cap set(dir(Dogma\_Genes))$
\FOR{$\text{gene in common}$}
        \STATE $gen1 \leftarrow open(NCBI\_Genes(gene)).read()$         
        \STATE $gen2 \leftarrow open(Dogma\_Genes(gene)).read()$
        \STATE $score \leftarrow geneChk(gen1,gen2)$
        \IF {$score > Threshold$}
                \STATE $geneList \leftarrow gene$
        \ENDIF 
\ENDFOR
\RETURN $geneList$
\end{algorithmic}
\end{algorithm}

\textbf{geneChk} is a subroutine used to find the best similarity score between 
two gene sequences after applying operations like \textit{reverse}, {\it complement}, 
and {\it reverse complement}. Algorithm~\ref{Alg3:genechk} gives the outline of 
geneChk subroutine.

\begin{algorithm}[H]
\caption{Find the Maximum Similarity Score between two sequences}
\label{Alg3:genechk}
\begin{algorithmic} 
\REQUIRE $gen1,gen2 \leftarrow \text{NCBI gene sequence, Dogma gene sequence}$
\ENSURE $\text{Maximum similarity score}$
\STATE $Score1 \leftarrow needle(gen1,gen2)$
\STATE $Score2 \leftarrow needle(gen1,Reverse(gen2))$
\STATE $Score3 \leftarrow needle(gen1,Complement(gen2))$
\STATE $Score4 \leftarrow needle(gen1,Reverse(Complement(gen2)))$
\RETURN $max(Score1, Score2, Score3, Score4)$
\end{algorithmic}
\end{algorithm}  

% THIS SUBSECTION MUST BE IMPROVED 

\subsubsection{Intersection Core Matrix (\textit{ICM})}

To extract  core genes, we  iteratively collect the maximum  number of
common  genes   between  genomes  and  therefore   during  this  stage
an \textit{Intersection  Core Matrix}  (ICM) is built.   ICM is  a two
dimensional symmetric matrix where each row and each column correspond
to   one   genome.   Hence,   an   element   of   the  matrix   stores
the  \textit{Intersection Score}  (IS):  the cardinality  of the  core
genes   set  obtained   by  intersecting   one  genome   with  another
one. Maximum  cardinality results in selecting the  two genomes having
the maximum score. Mathematically speaking, if we have $n$ genomes in
local database, the ICM is an $n \times n$ matrix whose elements
satisfy: 
\begin{equation}
score_{ij}=\vert g_i \cap g_j\vert
\label{Eq1}
\end{equation}
\noindent where $1 \leq i \leq n$, $1 \leq j \leq n$, and $g_i, g_j$ are 
genomes. The  generation of a new  core gene depends obviously on the
value of intersection scores $score_{ij}$:

% TO BE CONTINUED

$$
\text{new Core} = 
\begin{cases}
\text{Ignored} & \text{if $\textit{score}=0$;} \\
\text{new Core id} & \text{if $\textit{Score}>0$.}
\end{cases}
$$

if     $\textit{Score}=0$     then     we    have     \textit{disjoint
relation} \emph{i.e.},  no common genes between two  genomes.  In this
case  the  system  ignores  the   genome  that  annul  the  core  gene
size. Otherwise, The system removes these two genomes from ICM and add
new  core  genome  with a  \textit{coreID}  of  them  to ICM  for  the
calculation in  next iteration. This  process reduces the size  of ICM
and repeats until all genomes  are treated \emph{i.e.} ICM has no more
genomes.  We observe  that ICM is very large because  of the amount of
data that it stores. This results  to be time and memory consuming for
calculating  the  intersection  scores.   To  increase  the  speed  of
calculations, it  is sufficient to  only calculate the  upper triangle
scores. The time complexity for this process after enhancement is thus
$O(\frac{n.(n-1)}{2})$.   Algorithm   \ref{Alg1:ICM}  illustrates  the
construction of  the ICM matrix and  the extraction of  the core genes
where \textit{GenomeList},  represents the database  where all genomes
data are stored. At each iteration, it computes the maximum core genes
with its two genomes parents.

% ALGORITHM HAS BEEN REWRITTEN

\begin{algorithm}[H]
\caption{Extract Maximum Intersection Score}
\label{Alg1:ICM}
\begin{algorithmic} 
\REQUIRE $L \leftarrow \text{genomes vectors}$
\ENSURE $B1 \leftarrow Max Core Vector$ 
\FOR{$i \leftarrow 0:len(L)-1$}
        \STATE $score \leftarrow 0$
        \STATE $core1 \leftarrow set(GenomeList[L[i]])$
        \STATE $g1 \leftarrow L[i]$
        \FOR{$j \leftarrow i+1:len(L)$}
                \STATE $core2 \leftarrow set(GenomeList[L[j]])$
                \STATE $Core \leftarrow core1 \cap core2$
                \IF{$len(Core) > score$}
                  \STATE $score \leftarrow len(Core)$
                  \STATE $g2 \leftarrow L[j]$
                \ENDIF
        \ENDFOR
        \STATE $B1[score] \leftarrow (g1,g2)$
\ENDFOR
\RETURN $max(B1)$
\end{algorithmic}
\end{algorithm}

\subsection{Features visualization}

The goal is to visualize results  by building a tree of evolution. All
core  genes generated  represent  important information  in the  tree,
because they  provide information about  the ancestors of two  or more
genomes. Each  node in the  tree represents one chloroplast  genome or
one predicted core called \textit{(Genes count:Family name\_Scientific
names\_Accession number)}, while an edge is labeled with the number of
lost  genes from  a leaf  genome or  an intermediate  core  gene. Such
numbers are  very interesting because  they give an  information about
the evolution:  how many genes  were lost between two  species whether
they  belong  to  the  same  familie  or not.   By  the  principle  of
classification, a  small number of genes lost  among species indicates
that those species are close to  each other and belong to same family,
while a  large lost  means that we  have an  evolutionary relationship
between species  from different families. To depict  the links between
species   clearly,  we   built   a  phylogenetic   tree  showing   the
relationships based on the distances among genes sequences. Many tools
are    available   to    obtain    a   such    tree,   for    example:
PHYML\cite{guindon2005phyml},
RAxML{\cite{stamatakis2008raxml,stamatakis2005raxml},    BioNJ,    and
TNT\cite{goloboff2008tnt}}.    In   this  work,   we   chose  to   use
RAxML\cite{stamatakis2008raxml,stamatakis2005raxml}   because   it  is
fast, accurate,  and can build large  trees when dealing  with a large
number of genomic sequences.

The procedure used to built a phylogenetic tree is as follows:
\begin{enumerate}
\item For each gene in a core gene, extract its sequence and store it in the database.
\item Use multiple alignment tools such as (****to be write after see christophe****) 
to align these sequences with each others.
\item Submit the resulting aligned sequences to RAxML program to compute the distances and finally draw the phylogenetic tree.
\end{enumerate} 

\begin{figure}[H]
  \centering \includegraphics[width=0.8\textwidth]{Whole_system}
  \caption{Overview of the pipeline}\label{wholesystem}
\end{figure}

\section{Implementation}
We implemented four algorithms to extract maximum core genes from large amount of chloroplast genomes. Two algorithms used to extract core genes based on NCBI annotation, and the others based on dogma annotation tool. Evolutionary tree generated as a result from each method implementation. In this section, we will present the four methods, and how they can extract maximum core genes?, and how the developed code will generate the evolutionary tree.

\subsection{Extract Core Genes based on Gene Contents}

\subsubsection{Core Genes based on NCBI Annotation}
The first idea to construct the core genome is based on the extraction of Genes names (as gene presence or absence). For instant, in this stage neither sequence comparison nor new annotation were made, we just want to extract all genes with counts stored in each chloroplast genome, then find the intersection core genes based on gene names. \\
The pipeline of extracting core genes can summarize in the following steps according to pre-processing method used:\\

\begin{enumerate}
\item We downloads already annotated chloroplast genomes in the form of fasta coding genes (\emph{i.e.} \textit{exons}).
\item Extract genes names and apply to solve gene duplication using first method. 
\item Convert fasta file format to geneVision file format to generate ICM. 
\item Calculate ICM matrix to find maximum core \textit{Score}. New core genes for two genomes will generate and a specific \textit{CoreId} will assign to it. This process continue until no elements remain in the matrix.
\item Evolutionary tree will take place by using all data generated from step 1 and 4. The tree will also display the amount of genes lost from each intersection iteration. A specific excel file will be generated that store all the data in local database.
\end{enumerate}

There main drawback with this method is genes orthography (e.g two different genes sequences with same gene name). In this case, Gene lost is considered by solving gene duplication based on first method to solve gene duplication.

\subsubsection{Core Genes based on Dogma Annotation}  
The main goal is to get as much as possible the core genes of maximum coding genes names. According to NCBI annotation problem based on \cite{Bakke2009}, annotation method like dogma can give us more reliable coding genes than NCBI. This is because NCBI annotation can carry some annotation and gene identification errors. The general overview of whole process of extraction illustrated in figure \ref{wholesystem}.

extracting core genes based on genes names and counts summarized in the following steps:\\
\begin{enumerate}
\item We apply the genome annotation manually using Dogma annotation tool. 
\item Analysing genomes to store lists of code genes names (\textit{i.e. exons}). solve gene fragments is done by using first method in solve gene fragments. The output from annotation process with dogma is genomes files in GenVision file format. Sets of genes were stored in the database.
\item Generate ICM matrix to calculate maximum core genes.
\item Draw the evolutionary tree by extracted all genes sequences from each core. Then applying multiple alignment process on the sequences to calculate the distance among cores to draw a phylogenetic tree.

\end{enumerate}


The main drawback from the method of extracting core genes based on gene names and counts is that we can not depending only on genes names because of three causes: first, the genome may have not totally named (This can be found in early versions of NCBI genomes), so we will have some lost sequences. Second, we may have two genes sharing the same name, while their sequences are different. Third, we need to annotate all the genomes.               

\subsection{Extract Core Genes based on Gene Quality Control}
The main idea from this method is to focus on genes quality to predict maximum core genes. By comparing only genes names from one annotation tool is not enough. The question here, does the predicted gene from NCBI is the same gene predicted by Dogma based on gene name and gene sequence?. If yes, then we can predict new quality genomes based on quality control test with a specific threshold. Predicted Genomes comes from merging two annotation techniques. While if no, we can not depending neither on NCBI nor Dogma because of annotation error. Core genes can by predicted by using one of the 

\subsubsection{Core genes based on NCBI and Dogma Annotation}
This method summarized in the following steps:\\

\begin{enumerate}
\item Retrieve the annotation of all genomes from NCBI and Dogma: in this step, we apply the annotation of all chloroplast genomes in the database using NCBI annotation and Dogma annotation tool.
\item Convert NCBI genomes to GeneVision file format, then apply the second method of gene defragmentation methods for NCBI and dogma genomes.  
\item Predict quality genomes: the process is to pick a genome annotation from two sources, extracting all common genes based on genes names, then applying Needle-man wunch algorithm to align the two sequences based on a threshold equal to 65\%. If the alignment score pass the threshold, then this gene will removed from the competition and store it in quality genome by saving its name with the largest gene sequence with respect to start and end codons. All quality genomes will store in the form of GenVision file format.    
\item Extract Core genes: from the above two steps, we will have new genomes with quality genes, ofcourse, we have some genes lost here, because dogma produced tRNA and rRNA genes while NCBI did not generate rRNA genes and vise-versa. Build ICM to extract core genes will be sufficient because we already check genes sequences.
\item Display tree: An evolution tree then will be display based on the intersections of quality genomes.   
\end{enumerate}