Description Usage Arguments Details Value References See Also Examples

View source: R/estimate_function_effects.R

Given within topic functional predictions, estimate the effects at a given gene function category level. The effects correspond to a topic-gene category interaction term after accounting for topic and gene category effects. The model can be fit via either maximum likelihood or Hamiltonian MC.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 | ```
## S3 method for class 'functions'
est(
object,
topics_subset,
level = 2,
method = c("hmc", "ml"),
seed = object$seeds$next_seed,
verbose = FALSE,
...
)
## S3 method for class 'hmc'
est(
object,
inits,
prior = c("t", "normal", "laplace"),
t_df = c(7, 7, 7),
iters = 300,
warmup = iters/2,
chains = 1,
cores = 1,
seed = sample.int(.Machine$integer.max, 1),
return_summary = TRUE,
verbose = FALSE,
...
)
## S3 method for class 'ml'
est(
object,
iters = 1000,
verbose = FALSE,
seed = sample.int(.Machine$integer.max, 1),
...
)
``` |

`object` |
(required) Ouput of |

`topics_subset` |
Vector of topic indexes to be evaluated. Recommended to be < 25. |

`level` |
Gene category level to evaluate. Defaults to 2. |

`method` |
String indicating either ml or hmc. Defaults to hmc. |

`seed` |
Seed for the random number generator to reproduce previous results. |

`verbose` |
Logical flag to print progress information. Defaults to FALSE. |

`...` |
Additional arguments for methods. |

`inits` |
List of values for parameter initialization. If omitted, values
are generated via |

`prior` |
Prior to be placed on covariate weights. Choices include student-t, normal, and laplace. Defaults to student-t. |

`t_df` |
Degrees of freedom for student-t priors. Defaults to 7. |

`iters` |
Number of iterations for for fitting. Defaults to 300 and 100 for HMC and ML, respectively. |

`warmup` |
For HMC, proportion of iterations devoted to warmup. Defaults to iters/2. |

`chains` |
For HMC, number of independent chains. Defaults to 1. |

`cores` |
For HMC, number of cores to parallelize chains. Defaults to 1. |

`return_summary` |
Logical flag to return results summary. Defaults to TRUE. |

The functional effects are estimated via a multilevel Bayesian negative binomial regression model. Topic and pathway level effects are estimated, as well as topic-pathway interactions. The model has the following form:

*θ_{i} = μ + β_{w} + β_{k} + β_{w,k}*

*y_{i} ~ NB(θ_{i},φ)*

where *μ* is the intercept and each *β* term represents the weight for pathway level,
topic, and pathway level-topic interaction, respectively; *φ* represents the dispersion
parameter.

Hamiltonian MC is performed via Stan. By default, student-t priors with degrees of
freedom set at 7 are placed on all regression weights, with variance terms distributed by half normal
priors. The intercept *μ* is given a normal prior with fixed variance. Lastly, *φ* is
given an *exponential(.5)* prior. The priors placed on the regression weights can be changed by
the user to either normal, t-family, or laplace (double exponential) priors if a sparse solution is
desired. For the latter, each variance term is given an additional regularization parameter
*λ* which in turn is distributed by a *chi-squared(1)* distribution.

Unless a set of initialization values are provided by the user, or the user chooses to select a random
initialization procedure, initial values are set at the maximum likelihood estimate via
`glmer.nb`

, but at a far smaller number of iterations than had the user chosen
ML as his or her estimation method.

Maximum likelihood estimation is performed via `glmer.nb`

. For
deeper level functional categories, the model may fail to converge, even with a substantial
number of iterations. In such a case, the model estimates are returns so the user can perform
HMC, but by initializing at these ML values.

An object of class effects containing

- model
List containing the parameters, fit, and summary.

- gene_table
Dataframe containing the formatted predicted gene information from

`predict.topics`

.

Bates, D., Maechler, M., Bolker, B., and Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48. doi:10.18637/jss.v067.i01.

Gelman, A. and Hill, J. (2006). Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press; 1 edition.

Stan Development Team. 2016. RStan: the R interface to Stan. http://mc-stan.org

Stan Development Team. 2016. Stan Modeling Language Users Guide and Reference Manual, Version 2.14.0. http://mc-stan.org

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | ```
formula <- ~DIAGNOSIS
refs <- 'CD'
dat <- prepare_data(otu_table=GEVERS$OTU,rows_are_taxa=FALSE,tax_table=GEVERS$TAX,
metadata=GEVERS$META,formula=formula,refs=refs,
cn_normalize=TRUE,drop=TRUE)
## Not run:
topics <- find_topics(dat,K=15)
functions <- predict(topics,reference_path='/references/ko_13_5_precalculated.tab.gz')
function_effects <- est(functions,level=3,
iters=500,method='hmc',
prior=c('laplace','t','laplace'))
## End(Not run)
``` |

Embedding an R snippet on your website

Add the following code to your website.

For more information on customizing the embed code, read Embedding Snippets.