from the ISB Microbiome Course 2023
First, we'll need the slides, full of digestible information
https://gibbons-lab.github.io/isb_course_2023/16S
📕 Let's switch to the notebook and get started
Click me to open the notebook!
All output we generate can be found in the treasure_chest folder at
https://github.com/Gibbons-Lab/isb_course_2023/tree/main/treasure_chest
or materials/treasure_chest in the Colaboratory notebook.
- 30-40 trillion bacterial cells
- Heterogenous between individuals
- Helps digest food and produces metabolites
- Affects our health
How do we see what is in the microbiome?
- Hundreds to Thousands of taxa in each person
- Difficult to culture outside the resident environment
- We can sequence their DNA instead
Pronounced like chime 🔔
Created ~2010 during the Human Microbiome Project (2007 - 2016) under the leadership of Greg Caporaso and Rob Knight.
QIIME 2 is a powerful, extensible, and decentralized microbiome analysis package with a focus on data processing and analysis transparency.
Quantitative insights into Microbial Ecology
Essentially, QIIME is a set of commands to transform microbiome data into intermediate outputs and visualizations.
It's commonly used via the command line. We'll use it within the Colab Notebook
QIIME 2 comes with a lot of help, including a wide range of tutorials, general documentation and a user forum where you can ask questions.
QIIME 2 manages artifacts, which are basically intermediate data that feed into actions to either produce other artifacts or visualizations.
Artifacts often represent intermediate steps, but Visualizations are end points meant for human consumption ☝️.
Visualizations cannot be used as inputs for additional commands
The 16S gene is universal and contains interspersed conserved regions perfect for PCR priming and hypervariable regions with phylogenetic heterogeneity.
16S amplicon sequencing data of the V4 region from human fecal samples
4 healthy donors and 4 individuals with recurrent infection.
https://doi.org/10.1186/s40168-015-0070-0
To start, we'll import our raw data into QIIME as an artifact.
💻 Let's switch to the notebook and get started
We just ran the DADA2 plugin for QIIME, which is doing 4 things:
- filter and trim the reads
- find the most likely original sequences in the sample (ASVs)
- remove chimeras
- count the abundances
- trim low quality regions
- remove reads with low average quality
- remove reads with ambiguous bases (Ns)
- remove PhiX (added to sequencing)
Expectation-Maximization (EM) algorithm to find amplicon sequence variants (ASVs) and the real error model at the same time.
The primers used in this study were F515/R806. How long is the amplified fragment?
We now have a table containing the counts for each ASV in each sample. We also have a list of ASVs.
:thinking_face: Do you have an idea for what we could do with those two data sets? What quantities might we be interested in?
In microbial community analysis we are usually interested in two different families of diversity metrics, alpha diversity (ecological diversity within a sample) and beta diversity (ecological differences between samples).
How diverse is a single sample?
- richness: how many taxa do we observe (richness)?
→ #observed taxa - evenness: how evenly are abundances distributed across taxa?
→ Evenness index - mixtures: metrics that combine both richness and evenness
→ Shannon Index, Simpson's Index
Alpha diversity will provide a single value for each sample.
It can be treated as any other sample measurement and is suitable for classic univariate tests (t-test, Mann-Whitney U test).
How different are two or more samples/donors/sites from one another other?
- unweighted: how many taxa are shared between samples?
→ Jaccard index, unweighted UniFrac - weighted: do shared taxa have similar abundances?
→ Bray-Curtis distance, weighted UniFrac
Do samples share genetically similar taxa?
Weighted UniFrac scales branches by abundance.
One of the basic things we might want to look at is how the sequences across all samples are related to one another. That is, we are often interested in their phylogeny.
Phylogenetic trees are built from multiple sequence alignments and sequences are arranged by sequence similarity (branch length).
More complicated. Usually not normal and very heterogeneous. PERMANOVA can deal with that.
💻 Let's switch to the notebook and calculate the diversity metrics
We are still just working with sequences and have no idea what organisms they correspond to.
:thinking_face: What would you do to go from a sequence to an organism's name?
Even though directly aligning our sequences to a database of known genes seems most intuitive, this does not always work well in practice. Why?
Instead, use subsequences (k-mers) and their counts to predict the lineage/taxonomy with machine learning methods. For 16S amplicon fragments this often provides better generalization and faster results.
💻 Let's switch to the notebook and assign taxonomy to our ASVs
Which taxa are associated with the disease state?
Alex Carr
Sean Gibbons
Alyssa Easton
Katherine Ramos Sarmiento
Noa Rappaport
Karl Gaisser
Chloe Herman
Greg Caporaso
Christian Diener
Dominic Lewis
Allison Kudla
Audri Hubbard
Joe Myxter
Thea Swanson
Victoria Uhl
Connor Kelly
Shanna Braga
ISB Facilities Team