Help with Temporal Signal

[Ed Djo]

Hi everyone,

I’m currently evaluating the temporal signal in a subclade dataset of E. coli genomes (~800 samples) before deciding whether to proceed with a BEAST analysis. The samples range from 1998 to 2025. The tree was inferred using IQ-TREE.

I’ve assessed the tree using TempEst, and I noticed the following:

-Without selecting the best-fitting root, I get a positive slope of 1.48e-5, with an R² = 0.0069 and a correlation coefficient = 0.0818.

-With the best-fitting root enabled, the slope becomes negative (-0.000), the R² drops to 0.0217, and the correlation coefficient = -0.1474

I’ve already:

-Removed clear outliers flagged in TempEst

-Tried trimming the dataset to achieve more uniform temporal sampling.

-Used decimal-formatted sample dates.

Is there anything else I can do to verify or enhance the temporal signal? Or should I interpret this as evidence that my dataset may not be suitable for BEAST dating?

Appreciate any advice or experience others can share.

Thanks in advance,

Ed

[Gaspary Eugene]

Dear Ed,

Completely there is no sufficient signal to proceed with BEAST it will not converge....will stuck forever

I have some questions 

1. What target is this sequence....16s...or these all from same target?

2. Looks there is quite disperse sampling times .....root to tip distance give sampling times is certain 

3. When you fit the the tree does current tips appear on top of the tree ? When it is becoming worse after fitting 

My suggestions 

1. You can remove outliers based on sampling times 

2. Estimate Molecular clock using treeTime and examine residuals on the tree

I seee 

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/beast-users/3c1f18ad-8ebd-4d6c-bde2-886b495b0e86n%40googlegroups.com.

[Artem B]

Hi Ed,

You shouldn't remove outliers just because they are outliers, this forces data to a model (strict clock) but we want the opposite. There should be a reason to remove outliers (date mislabeling,  contamination, sequencing errors, etc.). The negative slope may indicate a high degree of among-lineage rate variation implying a relaxed clock. You can try to explore temporal signal in the fully Bayesian test (BETS). See my answer here: https://groups.google.com/g/beast-users/c/vjm9iLJEUeI/m/RxsH2LWIBgAJ

Best,
Artem

суббота, 26 июля 2025 г. в 04:37:29 UTC+8, Ed Djo: