What can I do with BEAM?
BEAM is a probability model of amino-acid or nucleotide sequences, learned from a collection of sequences and, if you have them, their abundances. The paper and the examples on this site use T cell receptor β (TCRβ) CDR3s, but the same questions apply to any collection of sequences you can build a model from. Find your question below; each links to a worked example you can run.
Single sequences
Section titled “Single sequences”| Question | Where |
|---|---|
| What is the probability of this sequence, for example a TCR CDR3? | How likely is a sequence? |
| Which residues make a sequence likely or unlikely? | Which residues drive a score |
| How do I generate realistic synthetic sequences, or a background set? | Generating sequences |
| Which group, cell type or condition does this sequence look like? | Comparing groups |
Sequencing depth, diversity and overlap
Section titled “Sequencing depth, diversity and overlap”| Question | Where |
|---|---|
| How deeply should I sequence? How many distinct sequences (clonotypes) will I see? | How deeply should I sequence? |
| Have I sequenced enough? How likely is the next cell to be new? | How diverse is my sample? |
| How biased is the Shannon entropy (diversity) I measure in a sample? | How diverse is my sample? |
| How many sequences will two samples share by chance? | How many sequences will two samples share? |
Public and private sequences
Section titled “Public and private sequences”| Question | Where |
|---|---|
| How many people will carry this sequence? Is it public? | How public is a sequence? |
| Which sequences of my repertoire are likely to be public? | The most public sequences of a repertoire |
| How many sequences in a cohort are private, and how many shared? | The publicness of a whole cohort |
| What do public sequences look like? | What public sequences look like |
Comparing repertoires, cell types and conditions
Section titled “Comparing repertoires, cell types and conditions”| Question | Where |
|---|---|
| How different are two repertoires, and why? | How different are two repertoires? |
| Which sequence patterns differ between two repertoires? | Reading the patterns |
| How do I compare cell types (CD4/CD8, naive/memory), tissues or patient groups? | Comparing groups |
| How do I cluster many repertoires, or find a sample’s nearest neighbours? | Clustering repertoires |
Building models
Section titled “Building models”| Question | Where |
|---|---|
| How do I build a model from my repertoire or an AIRR file? | Building a model from your data |
| Should I build one model per repertoire, per group, or one reference? | Which model do I need? |
| Should I weight by abundance, or count each sequence once? | Abundance or unique sequences? |
| Does my model describe my data well? | Checking a model |
| How do I keep my own cohort out of the reference? | Leaving a cohort out |
| How does BEAM work? | How BEAM works |