ababaian
commented
4 years ago One important note which has come up in going over the CoV+ classification data is that many "single-cell RNAseq" libraries are yielding CoV/Viral+ hits but have very poor coverage over the genome for the virus itself. It appears that this is a combination of high-error rate and PCR-duplications which is causing this.
SRR10338100
SUMZER_COMMENT=sra=SRR10338100,genome=cov3ma,date=200607-11:24;
family=Poxviridae;score=56;pctid=100;aln=111;glb=47;panlen=181133;cvg=_.oo..oOO._.o..____.o.O.O;top=NC_024446.1;topaln=15;toplen=306862;topname=Penguinpox virus isolate PSan92;
family=Herpesviridae;score=44;pctid=100;aln=104;glb=16;panlen=148687;cvg=__O._oo__.__O__Oo_oO.O__O;top=NC_002665.1;topaln=5;toplen=108873;topname=Bovine herpesvirus 4 long unique region;
family=Coronaviridae;score=40;pctid=100;aln=99;glb=6;panlen=30000;cvg=._.O__...__O..Oo.o___oO._;top=JX548304.1;topaln=6;toplen=13216;topname=Shorebird coronavirus isolate 85 RNA polymerase gene, partial cds;
...
acc=NC_006998.1;pctid=99.9;aln=45;glb=44;len=194711;cvgpct=24;len=194711;depth=0.0223;cvg=_ooOo.OO__o._.___..O_Ooo.;fam=Poxviridae;name=Vaccinia virus;
acc=NC_014969.1;pctid=100.0;aln=4;glb=0;len=44055;cvgpct=4;len=44055;depth=0.00209;cvg=________________________O;fam=Adenoviridae;name=Fowl adenovirus E;
acc=NC_038276.1;pctid=100.0;aln=4;glb=0;len=10881;cvgpct=4;len=10881;depth=0.00846;cvg=___________O_____________;fam=Rhabdoviridae;name=Nishimuro virus viral cRNA for hypothetical proteins;
acc=MG764156.1;pctid=100.0;aln=3;glb=0;len=13195;cvgpct=4;len=13195;depth=0.00523;cvg=O________________________;fam=Coronaviridae;name=Avian coronavirus isolate CoV-9698-2016/11/28-BP/SA orf1ab polyprotein (orf1ab) gene, partial cds;
acc=MG916904.1;pctid=100.0;aln=7;glb=0;len=28751;cvgpct=4;len=28751;depth=0.0056;cvg=______________O__________;fam=Coronaviridae;name=Bat coronavirus BtCoV/Rh/YN2012 isolate BtCoV/Rh/YN2012_Ra13591, complete genome;
acc=NC_036582.1;pctid=100.0;aln=4;glb=0;len=293123;cvgpct=4;len=293123;depth=0.000314;cvg=_____________________O___;fam=Poxviridae;name=Flamingopox virus FGPVKD09;
acc=AJ575397.1;pctid=100.0;aln=5;glb=0;len=20509;cvgpct=4;len=20509;depth=0.00585;cvg=O________________________;fam=Coronaviridae;name=Infectious bronchitis virus partial s1 gene for spike glycoprotein, S1 subunit, genomic RNA, isolate RF/17/98;
acc=NC_004065.1;pctid=100.0;aln=11;glb=0;len=230278;cvgpct=4;len=230278;depth=0.00114;cvg=__________________..___O_;fam=Herpesviridae;name=Murid herpesvirus 1;
acc=NC_020474.2;pctid=100.0;aln=4;glb=0;len=180421;cvgpct=4;len=180421;depth=0.000532;cvg=____O____________________;fam=Herpesviridae;name=Elephantid herpesvirus 1;
acc=NC_022613.1;pctid=100.0;aln=6;glb=0;len=43686;cvgpct=4;len=43686;depth=0.00316;cvg=____O____________________;fam=Adenoviridae;name=Turkey adenovirus 5 isolate 1277BT;
acc=NC_006879.1;pctid=100.0;aln=5;glb=0;len=34450;cvgpct=4;len=34450;depth=0.00334;cvg=_________________O_______;fam=Adenoviridae;name=Simian adenovirus 1;
acc=NC_001731.1;pctid=100.0;aln=7;glb=0;len=190289;cvgpct=4;len=190289;depth=0.000846;cvg=_________o_O_____________;fam=Poxviridae;name=Molluscum contagiosum virus subtype 1;
acc=NC_003310.1;pctid=100.0;aln=8;glb=0;len=196858;cvgpct=4;len=196858;depth=0.00098;cvg=____________________O____;fam=Poxviridae;name=Monkeypox virus Zaire-96-I-16;
acc=NC_027016.Negative filter of scRNAseq
I've added a scRNA-seq SraRunInfo.csv which we can use to seperate scRNAseq (which are likely no good for assembly) from other RNAseq libraries. This is available at s3://lovelywater/sra/scRNA_SraRunInfo.csv
rchikhi
rcedgar
Moving this discussion from slack. @ababaian suggested assembling 1k datasets for selected ranges of sumzer score, say 90-100, 80-89, 70-79 etc. with the goal of measuring how many of these datasets can be assembled in each score range. This would convert sumzer score to probability that a Cov can be assembled, which would inform a decision about where to set a score cutoff.
The summarizer score is designed to predict whether the dataset contains a given virus family. Whether this virus can be assembled is a different question.
If the goal is to predict assemblability, then a different score can be designed to take relevant factors into account, in particular read depth. Other relevant factors include read length, paired vs. unpaired, identity, and sequencing technology.
We don't have assemblers for long read/high error-rate technologies (I-T, PacBio) so these would score zero for now.
There is probably some minimum read length for minia & coronaspades, we should find out what this so that those datasets score zero.
If identity is low, then coverage will be underestimated due to low sensitivity of bowtie2.
With these points in mind, a first try at an assemblability score (a-score) could be:
assembly_score = coverage*depth_factor*identity_factor.The
depth_factorwould be a function which grows quickly from the minimum (depth=1) to a good depth (say,depth=3) and grows slowly fordepth>3. Theidentity_factorwould be 1 for high identities, then grow more quickly as the identity drops below 90% to compensate for the falling sensitivity of bowtie2, with values of say 3 or more below 80%. An a-score like this will be a better predictor of assemblability.