Iden%fica%on  of  long  non-­‐coding  RNAs  in  livestock  species    

   

Oana  Palasca  February  15th,  2018  

Overview  

     •  hGp://www.faang-­‐europe.org  •  Community  effort  to  establish  a  database  of  ncRNAs  in  

livestock  species    •  Annota%on  based  on  experimental  evidence  approach  (RNA-­‐

Seq)  •  Ini%ated  as  a  hackathon  in  October  2016      

2  

Background                  

Long  non  coding  RNAs    •  Capped,  polyadenylated,  alterna%vely  spliced  transcripts  •  Roles  in  regula%on  of  transcrip%on/transla%on,  chroma%n  modifica%on  

etc    Challenges  in  iden%fica%on  of  lncRNA  genes    

•  LiGle  sequence  conserva%on  –  difficult  to  predict  “de  novo”  •  Low  expression  levels  –  difficult  to  dis%nguish  from  transcrip%onal  noise  •  Low  consistency  between  biological  replicates  

3  

McMullen  et  al,  Clinical  Science,  2016  

Background  Current  annota%on  status,  Ensembl  91    

4  

0

5000

10000

human mouse pig cow sheep horse chicken

LincRNAs type

genestranscripts

0

50000

100000

150000

human mouse pig cow sheep horse chicken

Prot

ein

codi

ng

typegenestranscripts

Workflow  

Workflow  

 Data  collec%on  and  filtering    

 

 

       

1  

 Data  collec%on  and  filtering    

Ø  Data  obtained  from  the  European  Nucleo%de  Archive  

Ø  Cow,  horse,  pig,  sheep,  chicken  Ø  Selec%on  criteria:  Illumina,  paired-­‐ended,  

stranded,  >100  bp    

   

 

     

 

1  

Patro  et.  Al,  Nature  Methods,  2017  1  

 Data  collec%on  and  filtering    

Ø  Data  obtained  from  the  European  Nucleo%de  Archive  

Ø  Cow,  horse,  pig,  sheep,  chicken  Ø  Selec%on  criteria:  Illumina,  paired-­‐ended,  

stranded,  >100  bp    

Ø  Quality  check  –  FastQC  Ø  Strandedness  check  –  Salmon      

Ø  15%  samples  unstranded    Ø  10%  samples  with  first  read  mapping  to  

the  forward  strand      

 

     

 

1  

Patro  et.  Al,  Nature  Methods,  2017  1  

 Data  collec%on  and  filtering    

Ø  Data  obtained  from  the  European  Nucleo%de  Archive  

Ø  Cow,  horse,  pig,  sheep,  chicken  Ø  Selec%on  criteria:  Illumina,  paired-­‐ended,  

stranded,  >100  bp    

Ø  Quality  check  –  FastQC  Ø  Strandedness  check  –  Salmon      

Ø  15%  samples  unstranded    Ø  10%  samples  with  first  read  mapping  to  

the  forward  strand    

Ø  ~  900  samples,  32  Brenda  %ssue  ontology  terms  

Ø  Muscle  and  brain  available  for  all  species  

 

 

   

   

1  

Patro  et.  Al,  Nature  Methods,  2017  1  

Workflow  

Comparison  of  pipelines      Star    /Cufflinks    (SC)  vs.  Hisat2/String%e    (HS)  

   

         Input:    Ø  3  samples  from  chicken  (kidney,  liver,  heart  –  pooled)  Ø  5  samples  from  cow  (heart,  cerbral  cortex,  spleen,  liver,  kidney)  Ø  With/without  annota%on  

   

12  

     1    

Dobin  et.  al,  2013  1  

2  

Trapnell  et.al,  2010  2   3   Pertea  et.al,  2016  

3  

Comparison  of  pipelines  Total  number  of  transcripts,  per  %ssue  and  by  merging  all  %ssues    

   

         

   

Size  of  the  transcript  sets  obtained  by  running  the  two  pipelines,  COW  

0

25000

50000

75000

kidney liver heart merged

trans

crip

ts pipelineSC annotSC w/o annotHS annotHS w/o annot

0

20000

40000

60000

80000

heart c.cortex spleen liver kidney merged

trans

crip

ts pipelineSC annotSC w/o annotHS annotHS w/o annot

Size  of  the  transcript  sets  obtained  by  running  the  two  pipelines,  CHICKEN  

0

25000

50000

75000

kidney liver heart merged

trans

crip

ts pipelineSC annotSC w/o annotHS annotHS w/o annot

Comparison  of  pipelines  Impact  of  the  merging  step  string%e-­‐merge  on  the  SC  output  and  cuffmerge  on  the  HS  output  

   

         

   

Size  of  the  transcript  sets  obtained  by  running  the  two  pipelines,  COW  

0

25000

50000

75000

kidney liver heart merged merge_swap

trans

crip

ts pipelineSC annotSC w/o annotHS annotHS w/o annot

Size  of  the  transcript  sets  obtained  by  running  the  two  pipelines,  CHICKEN  

0

25000

50000

75000

kidney liver heart merged

trans

crip

ts pipelineSC annotSC w/o annotHS annotHS w/o annot

0

20000

40000

60000

80000

heart c.cortex spleen liver kidney merged merge_swap

trans

crip

ts pipelineSC annotSC w/o annotHS annotHS w/o annot

Comparison  of  pipelines  Recovery  rate  compared  to  annota%on  

   

         

   

Number  of  transcripts  exactly  overlapping  the  annota%on  

0

10

20

30

chicken cow%

from

pre

d. tr

ansc

ripts

pipelineSC w/o annotHS w/o annot

0

2000

4000

6000

chicken cow

num

ber o

f tra

nscr

ipts

pipelineSC w/o annotHS w/o annot

0

10

20

30

chicken cow

% fr

om p

red.

tran

scrip

ts

pipelineSC w/o annotHS w/o annot

Percentage  of  transcripts  exactly  overlapping  the  annota%on  from  the  total  transcripts  predicted  

Comparison  of  pipelines  Assessment  of  capacity  of  reconstruc%ng  lncRNAs  in  mouse    

Data  •  30  RNA-­‐Seq  mouse  samples  from  ENCODE/CSHL,  illumina,  paired-­‐ended,  100bp  •  Gencode  mouse  annota%on  down-­‐sampled  such  as  to  contain  a  set  of  genes  and  

transcripts  similar  to  e.g.  cow  (e.g.  20.000  PCGs  orthologous  with  cow  +  3000    randomly  selected  PCGs  +  2000  miRNas/snoRNAs)  

     

 

16  

Comparison  of  pipelines  Assessment  of  capacity  of  reconstruc%ng  lncRNAs  in  mouse    

Data  •  30  RNA-­‐Seq  mouse  samples  from  ENCODE/CSHL,  illumina,  paired-­‐ended,  100bp  •  Gencode  mouse  annota%on  down-­‐sampled  such  as  to  contain  a  set  of  genes  and  

transcripts  similar  to  e.g.  cow  (e.g.  20.000  PCGs  orthologous  with  cow  +  3000    randomly  selected  PCGs  +  2000  miRNas/snoRNAs)  

 Pipelines  to  test  •  STAR>Cufflinks-­‐>cuffmerge/string%e-­‐merge-­‐>FEELnc  •  HiSat2-­‐>String%e-­‐>string%e-­‐merge/cuffmerge-­‐>FEELnc          

 17  

Comparison  of  pipelines  Assessment  of  capacity  of  reconstruc%ng  lncRNAs  in  mouse    

Data  •  30  RNA-­‐Seq  mouse  samples  from  ENCODE/CSHL,  illumina,  paired-­‐ended,  100bp  •  Gencode  mouse  annota%on  down-­‐sampled  such  as  to  contain  a  set  of  genes  and  

transcripts  similar  to  e.g.  cow  (e.g.  20.000  PCGs  orthologous  with  cow  +  3000    randomly  selected  PCGs  +  2000  miRNas/snoRNAs)  

 Pipelines  to  test  •  STAR>Cufflinks-­‐>cuffmerge/string%e-­‐merge-­‐>FEELnc  •  HiSat2-­‐>String%e-­‐>string%e-­‐merge/cuffmerge-­‐>FEELnc    Output  •  Sensi%vity  and  specificity  for  predic%ng  lncRNAs  and  PCGs,  based  on  the  exact  

overlap  with  the  remaining  set  of  the  annota%on            

 

18  

Further  work  

•  Synteny  analysis  •  Correla%on  of  expression  levels  across  organisms  •  Structural  predic%ons  

•  Public  database,  connected  to  exis%ng  repositories,  e.g.  EBI        

 19  

Acknowledgements    

•  Daniel  Fischer  •  Sarah  Djebali  •  Chris%an  Anton  •  Alicja  Pacholewska  •  Nadezhda  Doncheva  •  Thomas  Derrien  •  Lel  Eory  •  Frank  Panitz  •  Magda  Mielczarek  •  Christa  Kuhn  •  Alan  Archibald  •  Jan  Gorodkin  

20