# Aggregate features using feature loadings This function aggregate the features into "meta features" by calculating a weighted summation of the features using feature loading of each component as weights. It can also aggregate features by using the combination of multiple components by ranking the features by a linear combination of feature loadings from multiple components. ```r aggregate_feature( res_tempted, mean_svd = NULL, datlist, pct = 1, contrast = NULL ) ``` ## Arguments - `res_tempted`: Output of `tempted`. - `mean_svd`: Output of `svd_centralize`. - `datlist`: Output of `format_tempted`, the original temporal tensor that will be aggregated. - `pct`: The percent of features to aggregate, features ranked by absolute value of the feature loading of each component. Default is 1, which means 100% of features are aggregated. Setting `pct=0.01` means top 1% of features is aggregated, where features are ranked in absolute value of feature loading of each component. - `contrast`: A matrix choosing how components are combined, each column is a contrast of length r and used to calculate the linear combination of the feature loadings of r components. ## Returns A list of results. - **metafeature_aggregate**: The meta feature obtained by aggregating the observed temporal tensor. It is a data.frame with four columns: "value" for the meta feature values, "subID" for the subject ID, "timepoint" for the time points, and "PC" indicating which component was used to construct the meta feature. - **metafeature_aggregate_est**: The meta feature obtained by aggregating the denoised temporal tensor. It has the same structure as `metafeature_aggregate`. - **contrast**: The contrast used to linearly combine the components from input. - **toppct**: A matrix of TRUE/FALSE indicating which features are aggregated in each component and contrast. ## Examples ```r # Take a subset of the samples so the example runs faster # Here we are taking samples from the odd months sub_sample <- rownames(meta_table)[(meta_table$day_of_life%/%12)%%2==1] count_table_sub <- count_table[sub_sample,] processed_table_sub <- processed_table[sub_sample,] meta_table_sub <- meta_table[sub_sample,] datlist <- format_tempted(count_table_sub, meta_table_sub$day_of_life, meta_table_sub$studyid, pseudo=0.5, transform="clr") mean_svd <- svd_centralize(datlist, r=1) res_tempted <- tempted(mean_svd$datlist, r=2, smooth=1e-5) contrast <- matrix(c(1/2,1), 2, 1) res_aggregate <- aggregate_feature(res_tempted, mean_svd, datlist, pct=1, contrast=contrast) # plot the aggregated features group <- unique(meta_table[, c("studyid", "delivery")]) plot_metafeature(res_aggregate$metafeature_aggregate, group, bws=30) ``` ## References Shi P, Martino C, Han R, Janssen S, Buck G, Serrano M, Owzar K, Knight R, Shenhav L, Zhang AR. (2023) **Time-Informed Dimensionality Reduction for Longitudinal Microbiome Studies**. bioRxiv. doi: 10.1101/550749. [https://www.biorxiv.org/content/10.1101/550749](https://www.biorxiv.org/content/10.1101/550749).

aggregate_feature function