Local Similarity Analysis (LSA) in QIIME

[sdpapet]

Hello, I have some microbiome data and I collected these samples following a time series. I want to look into how microbial composition change across time.

 I would like to do a Local Similarity Analysis (LSA. I was wondering if QIIME support this analysis? If not, anyone can recommend any good software to do this.

[Jamie Morton]

Hi sdpapet,

I don't believe that this analysis is available in qiime.  However, judging from the abstract, I would be wary about trusting the results that come out of it (you can't use Pearson correlation with microbiome data).

If you want to do correlations, I'd take a look at SPEIC-EASI.  If you are ambitious, it maybe worthwhile trying to use balance trees.

Cheers,

Jamie

[sdpapet]

Hi Jamie,

Thanks. I am going to look at these papers.

Just to clarify with you that these two methods (SPEIC-EASI) or balance trees offering alternative ways to check the microbial composition across time, but their statistical methods are better? Right? Or these two are also LSA, but use different way to calculate correlation coefficient?

Thanks

[Jamie Morton]

I'm not too familiar with the LSA method, so I really can't speak for it.

Both of the methods that I mentioned are conceptually different.  SPEIC-EASI attempts to infer the underlying dependency graph by estimating the inverse covariance matrix.  

Balance trees doesn't calculate correlation, but performs a linear transformation to map species proportions to log ratios.  This allows for inference of subpopulations rather than species, while enable standard statistical approaches (linear models, mixed effects models, ...) to gain insights behind microbial community fluctuation.   Here's a quick 2min video giving the run down behind this approach in case you are interested.

Best,

Jamie