ndreey
commented
1 year ago
- How many samples do we require to make it statistically correct?
- We are interested in host-contamination impact.
Replicates of each host-contamination (HC) level will not be required as we want to compare the different HC levels against the control (0% HC). With that said, we do have a small sample size (1 control, 4 treated). Depending on the variability of the data and the magnitude of the effects, this could limit the statistical power of the analysis.
- If the magnitude of difference between control and HC is large and/or the variability is low. then the sample size of five can be sufficient to detect a statistical significance.
- If the magnitude of the difference is low and/or variability is high, then a larger sample size will be beneficial.
However, assumptions of normal distribution will be difficult to assume, therefore, non-parametric tests will be the chosen approach.
- Does the randomness of the abundances of the specific species matter when their endophytic group abundance is already set?
Not for the aim of determining the impact of host-contamination. The overall performance of the assembly and binning will be measured. The hypothesis is that the overall performance becomes worse with higher HC. The only important bin we are interested in is that of the host as we want to find a way to determine the host contigs and remove them.
We want to determine the difference between 0% host-contamination to X% host-contamination.
- Difference in assembly and genome binning performance.
- Control vs different levels of treatment.
Yes, and this will be determined by following the same benchmarking method that was presented by the second challenge of CAMI.
- Could be wise to determine the proportion of genomes that are unique or common using average nucleotide identity (ANI). This is because assembly and binners are affected by the ANI.
- UNIQUE: Genomes with < 95% ANI
- COMMON: Genomes with >= 95% ANI
- Assembly
- Evaluated using MetaQUAST
- MEGAHIT will be used as the assembler.
- metaHipMer was faster (2.1h to assemble 100GB) and performed slightly better but requires 2 TB memory.
- MEGAHIT took ca 10h but required the least amount of memory (42GB).
- MEGAHIT performance was very good. It was truly only on the common strains HipMer outshined everyone else.
- Binning
- Evaluated using AMBER.
- CONCOCT
- Ca 9h and 12GB RAM
- MetaBinner
- Ca 51h and 20GB RAM
- Vamb ???
- Did not score to well but the purity is interesting
- ca 6h and 13GB RAM
Plant vs gold standard assembly
Plant vs MEGAHIT assembly
We are also interested in how well the binner bins the host and how much the purity is of that bin.
- Because the aim is to identify the host bin and remove its contigs (or other filter method).
- So we can rerun metagenomics analysis without the host-contamination.
Purity and completeness will be very interesting metrics to determine how much other contigs would be removed if we remove that bin + how much host-contamination we might be missing.
It took 3hours to generate 10GB NGS data (19GB total) on my laptop (4 cores)