BCH709 Introduction to Bioinformatics: 15_RNA-Seq tutorial2

Course Evaluation

Students will have access to course evaluation You can log in with your NetID to http://www.unr.edu/evaluate and check live updating response rates for your course evaluations. Our institutional goal is to achieve an 85% response rate for all evaluations, and to help us achieve that, we rely on you as well as the students.

If we can achieve 100% response rate for evaluation, I will give you additional points for all of you.

Discussion is open

https://unr.canvaslms.com/courses/56453/discussion_topics/514795

The due date for the question is November 23rd 11:59pm The due date for discussion is December 4th 11:59pm

  1. Define the biological hypotheses or bottleneck you wish to address which is related to your research, state the approach of your experiment, also state your system, study organism, or study site, and provide justification for what is the goal of your biological hypotheses. Please provide enough background information that the other students can understand your biological hypotheses or bottleneck. If your experiments are complicated, consider briefly explaining the experimental design with reason. If you get more like will get points. (30 Points)

  2. Please provide the bioinformatics suggestion that you want to suggest for other people’s research hypotheses or bottleneck. It should be scientifically valid methods even if it does not exist. Provide enough information to create an experiment and if you want to create software, please provide reasons and explain what kind of software we need, which part of the hypotheses or bottleneck can be solved. If the software doesn’t exist, please provide the design or roadmap of your software. Citation is optional but recommended. Please provide an obstacle to other people’s suggestions. In addition, insights and addition will also get points. If you get more like will get points. (10 points per valid answer with reference or concept or hypothesis, a total of 70 Points, seven replies are needed )

3.  The suggestion needs to reply as threaded format.

Examples are below

Example

Tef (or Teff) is a warm season, C4-photosynthesis grass that is gaining popularity in the U.S. as a high-quality summer forage, fodder, and gluten-free grain. However, Tef has relatively tiny seeds compare to other C4 grass. Currently, the primary goal of my research is to determine the loci of seed color and size. We are trying to use Genome-wide Association Study (GWAS) https://en.wikipedia.org/wiki/Genome-wide_association_study) (Links to an external site.) to identify the locus of seed color and size. In our lab, we have 386 teff accession and all of them have different seed colors. We extracted all of 386 teff accession DNA and sequencing was done. But I don’t know how to check the size and colors. The phenotyping is the most important but the main bottleneck of our experiment. How can we facilitate this task?

Student A (This example answer will get 3 points) I cannot find any solution but you can use a similar approach such as colony counting. The accurate counting of plates with high numbers of CFUs is error-prone since it requires a high level of attention by the counter. In the microbiome and general biology field use colony count software to analyze whole plate count. The examples are below.

https://www.nature.com/articles/s41598-018-24916-9 (Links to an external site.)

http://opencfu.sourceforge.net/ (Links to an external site.)

Student B (This example answer will get 5 points) I found one software, especially for seed size and color. GrainScan software was designed for seed size and color estimation. GrainScan uses a grayscale image is derived from the scanned color image by converting Red and Green color channel averaging. Based on the grayscaled image, the dimension measurements will be provided which include area, perimeter, and surrogates for length and width the major and minor axes of the best fit ellipse. Another great point of this software will provide color measurements for each seed in CIELAB values based on user provide color calibration options.

https://plantmethods.biomedcentral.com/articles/10.1186/1746-4811-10-23 (Links to an external site.)

Student C (This example answer will get 5 points) I don’t know how to code python, but there are several image analysis packages such as scikit-image (Links to an external site.)

Base on scikit software, you can calculate circularity with “4 * pi * props.area / props.perimeter ** 2”

The props area can be calculated with number of pixels from the centroid approach. The axis location and length can be converted by using orientation value from props and axis value can be estimated cos(orientation) * length/ 2 and sin (orientation) * length/ 2.

Delete previous work

rm -rf  /data/gpfs/assoc/bch709-1/<YOURID>/rnaseq_assembly/trinity_out_dir

WORKTING PATH

mkdir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/
cd /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/

Check previous results

ll /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/ATH/bam



Conda environment

CONDA_INSTRUMENTATION_ENABLED=1 conda create -n BCH709 python=3.7

conda activate BCH709

CONDA_INSTRUMENTATION_ENABLED=1 conda install -y -c bioconda -c conda-forge  sra-tools minimap2 trinity star multiqc=1.9 samtools=1.9 trim-galore gffread seqkit kraken2


CONDA_INSTRUMENTATION_ENABLED=1 conda install -y -c bioconda -c conda-forge -c r openssl=1.0 r-base icu=58.2 bioconductor-ctc  bioconductor-deseq2=1.20.0 bioconductor-biobase=2.40.0  bioconductor-qvalue=2.16.0 r-ape  r-gplots r-fastcluster=1.1.25 libiconv

Conda Environment

conda activate BCH709

CONDA_INSTRUMENTATION_ENABLED=1 conda install -y -c bioconda -c conda-forge  sra-tools minimap2 trinity star multiqc=1.9 samtools=1.9 trim-galore gffread seqkit kraken2

Publication (Drosophila)

Ramond E et al., “Comparative RNA-Seq analyses of Drosophila plasmatocytes reveal gene specific signatures in response to clean injury and septic injury”, Plos one, 2019 June 29, 2020

SRA Bioproject site

https://www.ncbi.nlm.nih.gov/bioproject/PRJNA638422
Run ReleaseDate LoadDate spots bases spots_with_mates avgLength size_MB AssemblyName download_path Experiment LibraryName LibraryStrategy LibrarySelection LibrarySource LibraryLayout InsertSize InsertDev Platform Model SRAStudy BioProject Study_Pubmed_id ProjectID Sample BioSample SampleType TaxID ScientificName SampleName g1k_pop_code source g1k_analysis_group Subject_ID Sex Disease Tumor Affection_Status Analyte_Type Histological_Type Body_Site CenterName Submission dbgap_study_accession Consent RunHash ReadHash
SRR11968960 6/9/2020 17:10 6/9/2020 17:09 12256307 1237887007 0 101 378   https://sra-download.ncbi.nlm.nih.gov/traces/sra23/SRR/011688/SRR11968960 SRX8512716 4w1118-ci RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811237 SAMN15192434 simple 7227 Drosophila melanogaster w1118-ci-rep1         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 3D3A8EBF0A13F90F9305C5DD917E9AE2 A111523A7FB7106EE54D2D8337D2E8F2
SRR11968959 6/9/2020 17:09 6/9/2020 17:07 14144827 1428627527 0 101 432   https://sra-download.ncbi.nlm.nih.gov/traces/sra1/SRR/011688/SRR11968959 SRX8512717 5w1118-ci RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811238 SAMN15192435 simple 7227 Drosophila melanogaster w1118-ci-rep2         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 5515CADB5697C29CDC396F942C24F387 6D312D3B5BF5001309FF93CB968E584B
SRR11968958 6/9/2020 17:11 6/9/2020 17:09 16118803 1627999103 0 101 495   https://sra-download.ncbi.nlm.nih.gov/traces/sra60/SRR/011688/SRR11968958 SRX8512718 6w1118-ci RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811239 SAMN15192436 simple 7227 Drosophila melanogaster w1118-ci-rep3         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public FCC81714EB524E34632C58BDC1E4C162 9F494AF29716E7175EA1E4652B08F0B7
SRR11968957 6/9/2020 17:07 6/9/2020 17:05 6215784 627794184 0 101 188   https://sra-download.ncbi.nlm.nih.gov/traces/sra47/SRR/011688/SRR11968957 SRX8512719 7w1118-ec RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811240 SAMN15192437 simple 7227 Drosophila melanogaster w1118-ec-rep1         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 9FA82BA9A828F9BDDE839810689EFA4F CC558BFAAE5EE65BDD1CC1C690575F9D
SRR11968956 6/9/2020 19:58 6/9/2020 19:56 46628659 4709494559 0 101 1573   https://sra-download.ncbi.nlm.nih.gov/traces/sra59/SRR/011688/SRR11968956 SRX8512720 8w1118-ec RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811241 SAMN15192438 simple 7227 Drosophila melanogaster w1118-ec-rep2         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 16E3AE29FAC6BDFDB2B60F5300A02302 0F7770A244784C635FEC2DC814A1040C
SRR11968955 6/9/2020 17:13 6/9/2020 17:11 16299093 1646208393 0 101 496   https://sra-download.ncbi.nlm.nih.gov/traces/sra62/SRR/011688/SRR11968955 SRX8512721 9w1118-ec RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811242 SAMN15192439 simple 7227 Drosophila melanogaster w1118-ec-rep3         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public CFA33A602A41E07AC4EFBEED3D2A0FE3 4F5983B317885D4E8FFC4B3D312B7674
SRR11968964 6/9/2020 17:15 6/9/2020 17:12 22436848 2266121648 0 101 843   https://sra-download.ncbi.nlm.nih.gov/traces/sra49/SRR/011688/SRR11968964 SRX8512712 22w1118-l3 RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811233 SAMN15192443 simple 7227 Drosophila melanogaster w1118-l3-rep1         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 2D2BB637C1817EC80B369D1EF0B39615 2136B5CFE75B7833A2A6927CF26E701E
SRR11968963 6/9/2020 19:33 6/9/2020 17:14 19826612 2002487812 0 101 740   https://sra-download.ncbi.nlm.nih.gov/traces/sra45/SRR/011688/SRR11968963 SRX8512713 23w1118-l3 RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811234 SAMN15192444 simple 7227 Drosophila melanogaster w1118-l3-rep2         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public A832FE389916D06C16C6F21DB93AD77A CE81D58F649BBBCECE551891816145AF
SRR11968962 6/9/2020 17:15 6/9/2020 17:12 20056763 2025733063 0 101 750   https://sra-download.ncbi.nlm.nih.gov/traces/sra11/SRR/011688/SRR11968962 SRX8512714 24w1118-l3 RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811235 SAMN15192445 simple 7227 Drosophila melanogaster w1118-l3-rep3         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 3F651F739352EAC0B28096237F2254EC 98BD1486600E785F9B5F8AC7DBCD4EA6
SRR11968954 6/9/2020 17:13 6/9/2020 17:10 16301608 1646462408 0 101 499   https://sra-download.ncbi.nlm.nih.gov/traces/sra20/SRR/011688/SRR11968954 SRX8512722 10w1118-sa RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811243 SAMN15192440 simple 7227 Drosophila melanogaster w1118-sa-rep1         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 55D22FA303406FAE40145D8A1E62598B 3E5BEB8C6FF03B853BA64D10989542E1
SRR11968966 6/9/2020 17:10 6/9/2020 17:08 16076977 1623774677 0 101 485   https://sra-download.ncbi.nlm.nih.gov/traces/sra50/SRR/011688/SRR11968966 SRX8512710 11w1118-sa RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811231 SAMN15192441 simple 7227 Drosophila melanogaster w1118-sa-rep2         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 9789FA28D07EBFD979E3DAE45E9D8CDF 54D9D5C5343EB9C9A7817434F1D4BB8B
SRR11968965 6/9/2020 17:10 6/9/2020 17:08 10379871 1048366971 0 101 316   https://sra-download.ncbi.nlm.nih.gov/traces/sra76/SRR/011688/SRR11968965 SRX8512711 12w1118-sa RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811232 SAMN15192442 simple 7227 Drosophila melanogaster w1118-sa-rep3         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 908A23B3924A405F6BE6D5362130E7B3 8BD0419DF27093A94546D02492F6661C
SRR11968968 6/9/2020 17:11 6/9/2020 17:09 16112703 1627383003 0 101 494   https://sra-download.ncbi.nlm.nih.gov/traces/sra51/SRR/011688/SRR11968968 SRX8512708 1w1118-uc RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811229 SAMN15192431 simple 7227 Drosophila melanogaster w1118-uc-rep1         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 39E07B4F04BC5A14AE664312E4DD5E67 27280649B15E4B766D86363C23679BE1
SRR11968967 6/9/2020 17:08 6/9/2020 17:06 9828233 992651533 0 101 302   https://sra-download.ncbi.nlm.nih.gov/traces/sra46/SRR/011688/SRR11968967 SRX8512709 2w1118-uc RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811230 SAMN15192432 simple 7227 Drosophila melanogaster w1118-uc-rep2         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public C9868075DE213901336D4DD9D22A9B72 558FD3E03B55FA60A231537D5E8EE198
SRR11968961 6/9/2020 17:15 6/9/2020 17:11 16343251 1650668351 0 101 498   https://sra-download.ncbi.nlm.nih.gov/traces/sra70/SRR/011688/SRR11968961 SRX8512715 3w1118-uc RNA-Seq Oligo-dT TRANSCRIPTOMIC SINGLE 0 0 ILLUMINA Illumina HiSeq 2500 SRP266662 PRJNA638422   638422 SRS6811236 SAMN15192433 simple 7227 Drosophila melanogaster w1118-uc-rep3         unknown   no         SWISS FEDERAL INSTITUTE OF TECHNOLOGY LAUSANNE SRA1085163   public 1725A91FA94755464378D8FF0F18A197 870D1C8B738F5A31C179B44124757B27

Fig 2. Transcriptome summaries from unchallenged whole larvae and hemocytes from unchallenged and infected larvae. (A) Transcriptome summary showing the number of reads for each triplicate in all experimental conditions with their corresponding number of mapped reads and the average percentage of alignment to the D. melanogaster genome. (B) Venn diagram representing the quantity of shared genes between all experimental treatments: Unchallenged wandering L3 larvae, hemocytes from unchallenged larvae, hemocytes from clean-pricked larvae (CI), hemocytes from larvae pricked with Escherichia coli (Ec), hemocytes from larvae pricked with Staphylococcus aureus (Sa).

Subset of data

Sample information Run
22w1118-l3 SRR11968964
23w1118-l3 SRR11968963
24w1118-l3 SRR11968962
10w1118-sa SRR11968954
11w1118-sa SRR11968966
12w1118-sa SRR11968965
cd /data/gpfs/assoc/bch709-1/wyim/RNA-Seq_example/
mkdir Drosophila && cd Drosophila

mkdir raw_data

mkdir trim

fastq-dump submission

#!/bin/bash
#SBATCH --job-name=fastqdump_Droso
#SBATCH --cpus-per-task=2
#SBATCH --time=2-15:00:00
#SBATCH --mem=16g
#SBATCH --mail-type=all
#SBATCH --mail-user=<youremail>
#SBATCH -o fastq-dump.out # STDOUT & STDERR
#SBATCH -p cpu-s2-core-0 
#SBATCH -A cpu-s2-bch709-1

fastq-dump  SRR11968964 --outdir ./raw_data  --gzip
fastq-dump  SRR11968963 --outdir ./raw_data  --gzip
fastq-dump  SRR11968962 --outdir ./raw_data  --gzip
fastq-dump  SRR11968954 --outdir ./raw_data  --gzip
fastq-dump  SRR11968966 --outdir ./raw_data  --gzip
fastq-dump  SRR11968965 --outdir ./raw_data  --gzip

Trim-galore

#!/bin/bash
#SBATCH --job-name=trim_Droso
#SBATCH --cpus-per-task=2
#SBATCH --time=2-15:00:00
#SBATCH --mem=16g
#SBATCH --mail-type=all
#SBATCH --mail-user=<youremail>
#SBATCH -o trim.out # STDOUT & STDERR
#SBATCH -p cpu-s2-core-0 
#SBATCH -A cpu-s2-bch709-1
#SBATCH --dependency=afterok:<PREVIOUS_JOBID>

trim_galore --cores 2  --max_n 40  --gzip -o trim raw_data/SRR11968964.fastq.gz --fastqc
trim_galore --cores 2  --max_n 40  --gzip -o trim raw_data/SRR11968963.fastq.gz --fastqc
trim_galore --cores 2  --max_n 40  --gzip -o trim raw_data/SRR11968962.fastq.gz --fastqc
trim_galore --cores 2  --max_n 40  --gzip -o trim raw_data/SRR11968954.fastq.gz --fastqc
trim_galore --cores 2  --max_n 40  --gzip -o trim raw_data/SRR11968966.fastq.gz --fastqc
trim_galore --cores 2  --max_n 40  --gzip -o trim raw_data/SRR11968965.fastq.gz --fastqc

Reference downloads

https://flybase.org/


cd /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila

mkdir bam

mkdir reference && cd reference

wget ftp://ftp.flybase.net/genomes/Drosophila_melanogaster/dmel_r6.36_FB2020_05/fasta/dmel-all-chromosome-r6.36.fasta.gz
wget ftp://ftp.flybase.net/genomes/Drosophila_melanogaster/dmel_r6.36_FB2020_05/gtf/dmel-all-r6.36.gtf.gz
wget ftp://ftp.flybase.net/genomes/Drosophila_melanogaster/dmel_r6.36_FB2020_05/fasta/dmel-all-CDS-r6.36.fasta.gz

gunzip dmel-all-chromosome-r6.36.fasta.gz
gunzip dmel-all-r6.36.gtf.gz
gunzip dmel-all-CDS-r6.36.fasta.gz

seqkit stats dmel-all-chromosome-r6.36.fasta
seqkit stats dmel-all-CDS-r6.36.fasta  

Reference index

#!/bin/bash
#SBATCH --job-name=reference_Droso
#SBATCH --cpus-per-task=2
#SBATCH --time=2-15:00:00
#SBATCH --mem=16g
#SBATCH --mail-type=all
#SBATCH --mail-user=<youremail>
#SBATCH -o trim.out # STDOUT & STDERR
#SBATCH -p cpu-s2-core-0 
#SBATCH -A cpu-s2-bch709-1

 STAR  --runThreadN 24 --runMode genomeGenerate --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --genomeFastaFiles  /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/dmel-all-chromosome-r6.36.fasta --sjdbGTFfile /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/dmel-all-r6.36.gtf  --sjdbOverhang 99 -genomeSAindexNbases 12

Mapping reads

#!/bin/bash
#SBATCH --job-name=mapping_Droso
#SBATCH --cpus-per-task=2
#SBATCH --time=2-15:00:00
#SBATCH --mem=16g
#SBATCH --mail-type=all
#SBATCH --mail-user=<youremail>
#SBATCH -o trim.out # STDOUT & STDERR
#SBATCH -p cpu-s2-core-0 
#SBATCH -A cpu-s2-bch709-1
#SBATCH --dependency=afterok:<trim_Droso>

STAR --runMode alignReads --runThreadN 8 --readFilesCommand zcat --outFilterMultimapNmax 10 --alignIntronMin 25 --alignIntronMax 100000 --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --readFilesIn /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/trim/SRR11968964_trimmed.fq.gz --outSAMtype BAM SortedByCoordinate --outFileNamePrefix /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/bam/SRR11968964.bam
STAR --runMode alignReads --runThreadN 8 --readFilesCommand zcat --outFilterMultimapNmax 10 --alignIntronMin 25 --alignIntronMax 100000 --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --readFilesIn /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/trim/SRR11968963_trimmed.fq.gz --outSAMtype BAM SortedByCoordinate --outFileNamePrefix /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/bam/SRR11968963.bam
STAR --runMode alignReads --runThreadN 8 --readFilesCommand zcat --outFilterMultimapNmax 10 --alignIntronMin 25 --alignIntronMax 100000 --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --readFilesIn /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/trim/SRR11968962_trimmed.fq.gz --outSAMtype BAM SortedByCoordinate --outFileNamePrefix /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/bam/SRR11968962.bam
STAR --runMode alignReads --runThreadN 8 --readFilesCommand zcat --outFilterMultimapNmax 10 --alignIntronMin 25 --alignIntronMax 100000 --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --readFilesIn /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/trim/SRR11968954_trimmed.fq.gz --outSAMtype BAM SortedByCoordinate --outFileNamePrefix /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/bam/SRR11968954.bam
STAR --runMode alignReads --runThreadN 8 --readFilesCommand zcat --outFilterMultimapNmax 10 --alignIntronMin 25 --alignIntronMax 100000 --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --readFilesIn /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/trim/SRR11968966_trimmed.fq.gz --outSAMtype BAM SortedByCoordinate --outFileNamePrefix /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/bam/SRR11968966.bam
STAR --runMode alignReads --runThreadN 8 --readFilesCommand zcat --outFilterMultimapNmax 10 --alignIntronMin 25 --alignIntronMax 100000 --genomeDir /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/reference/ --readFilesIn /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/trim/SRR11968965_trimmed.fq.gz --outSAMtype BAM SortedByCoordinate --outFileNamePrefix /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/bam/SRR11968965.bam

Investigate taxa

Here we introduce a software called Kraken2. This tool uses k-mers to assign a taxonomic labels in form of NCBI Taxonomy to the sequence (if possible). The taxonomic label is assigned based on similar k-mer content of the sequence in question to the k-mer content of reference genome sequence. The result is a classification of the sequence in question to the most likely taxonomic label. If the k-mer content is not similar to any genomic sequence in the database used, it will not assign any taxonomic label.

Donwload most recent database

https://benlangmead.github.io/aws-indexes/k2
wget https://genome-idx.s3.amazonaws.com/kraken/k2_standard_16gb_20200919.tar.gz
cd /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example

mkdir Kraken2 && cd Kraken2
kraken2-inspect --db  /data/gpfs/assoc/bch709-1/Course_material/database/  | head -5
```bash
#!/bin/bash
#SBATCH --job-name=Kraken_Droso
#SBATCH --cpus-per-task=2
#SBATCH --time=2-15:00:00
#SBATCH --mem=16g
#SBATCH --mail-type=all
#SBATCH --mail-user=<youremail>
#SBATCH -o trim.out # STDOUT & STDERR
#SBATCH -p cpu-s2-core-0
#SBATCH -A cpu-s2-bch709-1

kraken2  --threads 24 --report SRR11968954 --db /data/gpfs/assoc/bch709-1/Course_material/database/ /data/gpfs/assoc/bch709-1/<YOURID>/RNA-Seq_example/Drosophila/raw_data/SRR11968954.fastq.gz
"C"/"U": a one letter code indicating that the sequence was either classified or unclassified.
/data/gpfs/assoc/bch709-1/Course_material/kraken2_example

https://fbreitwieser.shinyapps.io/pavian/

Assignment1

cd /data/gpfs/assoc/bch709-1/wyim/RNA-Seq_example
multiqc . -n rnaseq1

Please upload rnaseq1.html to Webcampus.

Assignment2

Please run Kraken2 below six samples and generate Multiqc report for Kranken results only.

Sample information Run
22w1118-l3 SRR11968964
23w1118-l3 SRR11968963
24w1118-l3 SRR11968962
10w1118-sa SRR11968954
11w1118-sa SRR11968966
12w1118-sa SRR11968965