c***y
发帖数: 615 | 1 都是peptide sequences,分别试了neighbour joining, maximum likelihood, 还有
bayesian inference. 发现出来的结果完全是不一样的. 本人不是学进化的,只是在这
种情况下, 应该选哪个呢?为什么? | H*g
发帖数: 2333 | 2 I am not studying evolutionary biology neither. Just my 2cents. I guess I
would forget the neighbor-joining method.
Assuming that you have a good alignment with the most biological sense (
functional domain well aligned), good bootstrap value for ML and good
convergence for Bayesian Inference, regarding of the trees that are
generated from those different methods, you may have to distinguish them by
yourself by figuring out which one makes the most biological sense.
Without a thorough look at the trees, I tend to like the one by Bayesian
Inference, again assuming that you have a good alignment and a good
convergence.. | N******n
发帖数: 3003 | 3 what software does you use?
【在 c***y 的大作中提到】
: 都是peptide sequences,分别试了neighbour joining, maximum likelihood, 还有
: bayesian inference. 发现出来的结果完全是不一样的. 本人不是学进化的,只是在这
: 种情况下, 应该选哪个呢?为什么?
| c***y
发帖数: 615 | 4 My alignment came from clustalw
Bootstrapping NJ tree was generated using clustalx (100 bootstraps); ML tree
was generated using phylip package with PAM model; and BI tree was
generated using MrBayes...
Does software matter?
【在 N******n 的大作中提到】
: what software does you use?
| j********e
发帖数: 52 | 5 what's the bootstrapping scores for the three trees?
tree
【在 c***y 的大作中提到】
: My alignment came from clustalw
: Bootstrapping NJ tree was generated using clustalx (100 bootstraps); ML tree
: was generated using phylip package with PAM model; and BI tree was
: generated using MrBayes...
: Does software matter?
| N******n
发帖数: 3003 | 6 ML has a better statistical property,and inference the model parameter
statistically, but is very time consuming comparing to NJ.
Bayesian inference is become popular because of computational power
advancement. it considers the prior and the parameter together.
Only 2 cents from statistical perspective, not familiar with evolution.
【在 c***y 的大作中提到】
: 都是peptide sequences,分别试了neighbour joining, maximum likelihood, 还有
: bayesian inference. 发现出来的结果完全是不一样的. 本人不是学进化的,只是在这
: 种情况下, 应该选哪个呢?为什么?
| c***y
发帖数: 615 | 7 first need to correct some typo: bootstrapping NJ tree from 1000 bootstraps
instead of 100.
My running shows that BI tree takes longer time, 2 days; whereas ML tree
took less than 8 hrs.
【在 N******n 的大作中提到】
: ML has a better statistical property,and inference the model parameter
: statistically, but is very time consuming comparing to NJ.
: Bayesian inference is become popular because of computational power
: advancement. it considers the prior and the parameter together.
: Only 2 cents from statistical perspective, not familiar with evolution.
| c***y
发帖数: 615 | 8 how do I check the convergence for BI tree?
by
【在 H*g 的大作中提到】
: I am not studying evolutionary biology neither. Just my 2cents. I guess I
: would forget the neighbor-joining method.
: Assuming that you have a good alignment with the most biological sense (
: functional domain well aligned), good bootstrap value for ML and good
: convergence for Bayesian Inference, regarding of the trees that are
: generated from those different methods, you may have to distinguish them by
: yourself by figuring out which one makes the most biological sense.
: Without a thorough look at the trees, I tend to like the one by Bayesian
: Inference, again assuming that you have a good alignment and a good
: convergence..
| N******n
发帖数: 3003 | 9 the usual Bayesian approach obtains the posterior probability through MCMC
sampling and takes longer time. ML gets its parameter estimation through
numerical optimization and therefore, faster than BI.
bootstraps
【在 c***y 的大作中提到】
: first need to correct some typo: bootstrapping NJ tree from 1000 bootstraps
: instead of 100.
: My running shows that BI tree takes longer time, 2 days; whereas ML tree
: took less than 8 hrs.
| l**********1
发帖数: 5204 | 10 plus one article:
Algorithms Mol Biol. 2010 5:35.
Estimating the evidence of selection and the reliability of inference in
unigenic evolution.
by Fernandes AD et al,
HTTP: //www.ncbi.nlm.nih.gov/pubmed/21059250
and one Book
by Robert CP: The Bayesian choice: from decision-theoretic foundations to
computational implementation. 2 edition. Springer texts in statistics, New
York: Springer; 2001.
free down link:
HTTP: //ishare.iask.sina.com.cn/f/15300682.html | H*g
发帖数: 2333 | 11 For the convergence of Bayesian Inference tree, you could go to the .log
file, check the value behind the last "Average standard deviation of split
frequencies:". Without the info of this value at least, we would not know if
we have set the generation number or the probe temperature correctly, not
saying the resulted tree is appropriate for downstream analysis.
A good and biological meaningful alignment is essential for the analysis.
Make sure of this before jumping into the actual analysis. This means, if we
use the resulted alignment from ClustalW/MAFFT/T-COFFEE without any manual
adjustment, or if we haven't carefully examined the alignment from head to
toe, it is almost certain that the alignment is not ideal and the resulted
tree is destined to be inaccurate. Similarly, if we haven't compared the
alignment results from multiple alignment programs with different algorithms
, probably we haven't done enough homework to understand some basic concepts
at least, and are not ready to perform the analysis yet.
For the computation power, try an on-campus cluster or the free public
cluster CIPRES http://www.phylo.org/index.php/portal/. CIPRES has multiple built-in phylogenetic tools, including sequence alignment tools (e.g., CLustalX/MAFFT) and phylogenetic tree inference tools (e.g., RAxML and MrBayes).
I am not majored in phylogenetics. Just 2 cents.
【在 c***y 的大作中提到】
: how do I check the convergence for BI tree?
:
: by
| c***y
发帖数: 615 | 12 Thank you very much for your explanation. I am reading some relevant book
now, slowly though.
I checked my log file and got the following informatuiion:
Average standard deviation of split frequencies = 0.075483
Maximum standard deviation of split frequencies = 0.345682
Is it good or not?
Many articles mentioned manual adjustment after alignment. Does it mean I
just visually check and then do the adjustment? It sounds very time-
consuming and hard to be error-free since I have about 100 sequences and 900
aa in length...
if
we
manual
【在 H*g 的大作中提到】
: For the convergence of Bayesian Inference tree, you could go to the .log
: file, check the value behind the last "Average standard deviation of split
: frequencies:". Without the info of this value at least, we would not know if
: we have set the generation number or the probe temperature correctly, not
: saying the resulted tree is appropriate for downstream analysis.
: A good and biological meaningful alignment is essential for the analysis.
: Make sure of this before jumping into the actual analysis. This means, if we
: use the resulted alignment from ClustalW/MAFFT/T-COFFEE without any manual
: adjustment, or if we haven't carefully examined the alignment from head to
: toe, it is almost certain that the alignment is not ideal and the resulted
|
|
