This is the analysis of the tool Needle.
Please, download and install the tools: kallisto (v0.46.2), salmon(v1.5.1), REINDEER(v1.0.2) and Needle(v1.0.1). Then run the following commands.
For the accuracy analysis, you need the following two data sets: one data set is a simulated data set, the other one a real data set.
The simulated data set can be created by using the provided Rscript simulate.R.
Run:
Rscript simulate.R data/100.fa
The real data set is from the SEQC studies, more precisely, the data provided by Chinsanga et al. (https://github.com/ShiLab-Bioinformatics/GeneAnnotation) was analysed. In order to repeat the analysis, please download this data.
For the speed and space analysis, a well known data set containing breast, brain and blood RNA sequence files was used. The complete list
of experiments used by us can be found in data\large_dataset.lst. Please download these files as fastq.gz files.
The purpose of this analysis is to determine the accuracy of needle. In this analysis Needle is compared with the state-of-the-art tools kallisto and salmon as well as Reindeer.
The difference between the ground truth and the predicted expressions for the differential expression and the coverages
on the simulated data set can be obtained by running the following command. Please adaptt the paths in run_simulated.sh beforehand.
bash run_simulated.sh
This creates two Boxplots: Boxplot.png and Boxplot_Cov.png showing the estimated fold changes of the differential expression and the different coverages.
(It also creates Boxplot_03.png and Boxplot_Cov_03.png for a FPR of 0.3.)
The correlation values with the expression values of the RT-PCR and the microarray and the mean square error of the fold changes for all tools can be obtained by running the following command:
bash run_seqc.sh
The results are written in Results_SEQC_Data_set.txt (for a FPR of 0.3 in Results_SEQC_Data_set.txt).
The purpose of this analysis is to determine the space and speed consumption of Needle compared to its main competitor
REINDEER.
Time was measured by /usr/bin/time -v. (Tests were run with either 1 thread, 4 or 32 threads, add -t | --threads to the
commands to do the same.)
The data was then preprocessed and all k-mers with occurrences below a certain threshold were disregarded. (Thresholds
were based on the file size and follow the recommendation of Salmon et al.) The preprocess can be done by running the
scripts preprocess.sh and preprocess_reindeer.sh. (Please adjust the paths beforehand.)
bash preprocess.sh
preprocess_reindeer.sh data/samples.in
Then adjust the paths in run_large_dataset.sh and run:
bash run_large_dataset.sh
In the folders w_21, w_25, w_41 and reindeer the indexes and their measurement of /usr/bin/time -v for the construction and the queries can be found.
A summary of the results can be found in the file Results_Large_Data_set.txt containing a summary of the times, ram usages and sizes of the build and the query.
If you want the same analysis, run the scripts with same names and the _thread4 extension.
For this analysis, we need to know the names of the RNA-seq experiments and need to know, which tissue they are from. This information is provided in data/sras_1742.lst, data/brain_1742.lst, data/breast_1742.lst and data/blood_1742.lst.
Moreover, we need to estimate the expression of all proteincoding genes on a needle index, we picked here one we created under the space and speed analysis.
./needle estimate -i w_41/SRR_ data/gencode.v36.pc_transcripts.fa.gz -o gencode.out
Then you can run the following command to get all over expressed genes between one type of tissues compared to the other types according to t-test:
python3 ttest.py gencode.out data/sras_1742.lst data/breast_1742.lst data/blood_1742.lst data/brain_1742.lst
You can find the genes then in brain.lst, breast.lst and blood.lst, which you can then analyse with ShinyGo or any other gene ontology tool of your choice.
For this analysis, we copied the names of the breast cancer experiments from REINDEER to our repo (data/cancer_dataset.lst) and copied the list of experiments and their tissue of origin form here to our repo (data/sras_types.lst). With this information, we can check how well a distinction based on the quantification of the oncogenes CCND1, ERBB2,FOXM1 and MYC works by running the following command:
python3 oncogene_analysis.py data/cancer_dataset.lst gencode.out data/sras_1742.lst data/sras_types.lst