Paper Discussion of Glemin et al (2019)

Learning objectives for today

Understand what raw data the authors produced (type of data, individuals per species, sequencing platform) and how they transformed it into the alignments used for phylogenomic inference
Critically evaluate preprocessing/assembly/orthology decisions and how those choices can bias phylogenomic/hybridization inferences
Grasp the biological questions / systems (Aegilops/Triticum lineages, the A/B/D genomes) and which data features matter most for interpreting hybridization signals

Motivating questions

Framing the discussion

Q1: What do you think the authors are trying to convince us of in this paper, biologically, not methodologically?

Q2: What long-standing biological question about wheat evolution is this paper trying to answer?

Q3: What features of Aegilops/Triticum evolution make phylogenetic inference especially challenging?

Timescale of divergence?
Effective population sizes?
Hybridization compatibility?

What is the raw data?

Q4: What exactly did they measure? I.e., before any analysis, what is the biological object they actually sequenced?

Is this genomic DNA, RNA, or something else?
From how many individuals per species?
From what tissues? (does this matter?)

Q5: What biological information is missing by design? Given the data type and tissues sampled, are there any types of genes or signals that are guaranteed to be absent or under-represented?

Would every gene in the genome have an equal chance of appearing here?
Could two species differ biologically but appear similar because of expression?
How might this matter for detecting hybridization?

Q6: Given these limitations, why might the authors have chosen transcriptomes instead of whole genomes?

Cost?
Genome size?
Orthology vs paralogy?
Computational tractability at the time?

From reads to genes—preprocessing decisions

Q7: What is the first irreversible decision in the pipeline? Where in the pipeline do the authors make the first choice that permanently discards data?

Trimming thresholds?
Discarding short CDS?
Removing ambiguous cluster assignments?

Q8: Given the filters they use (length cutoffs, clustering rules), what kinds of genes are more likely to be removed?

Short genes?
Fast-evolving genes?
Recently duplicated genes?

Q9: Is their orthology strategy conservative or aggressive? They only keep CDS that match exactly one cluster bait. Is this a conservative or aggressive choice — and conservative with respect to what?

Conservative against false orthology or false paralogy?
Which biological scenarios suffer under this rule?
How might this affect inference of hybridization vs ILS?

Q10: If you changed one preprocessing decision, which would it be? Which preprocessing step would you most want to sensitivity-test, and why?

Length cutoff?
Identity threshold?
Use of consensus sequences later?

Gene trees, conflict, and biological interpretation

Q11: How do the authors decide it’s hybridization and not “just ILS”? What kind of evidence would you need to draw this conclusion?

Q12: What conclusions are most sensitive to the data choices? Which parts of the biological story would you trust the least if the data preprocessing were slightly different?

Presence vs absence of hybridization?
Identity of parental lineages?
Timing/order of events?
Others?

Summary

Q13: If you had to summarize this paper in one sentence that mentions both biology and data limitations, what would it be?