, Ai-Ling Lin [aut],
Jianlin Cheng [aut]| Title: | Weighted Omics View Embedding via Nystrom for Incomplete Multi-Omics Data |
|---|---|
| Description: | Supervised multi-omics integration for block-missing ("ragged") data. WOVEN learns a shared latent space across V omics modalities using only fully-observed anchor subjects, then projects block-missing subjects via Nystrom extension without feature-level imputation. Supervision via label-augmented cross-covariance (analogous to DIABLO) with optional sparse projection matrices (PMD). Designed for comparative effectiveness research where intersection-only methods introduce selection bias. |
| Authors: | Nathan Bresette [aut, cre]
, Ai-Ling Lin [aut],
Jianlin Cheng [aut] |
| Maintainer: | Nathan Bresette <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 0.99.0 |
| Built: | 2026-06-24 20:36:43 UTC |
| Source: | https://github.com/BiocStaging/woven |
Plots the first two latent dimensions from fit$Z, colored by group
label. Anchor subjects (complete cases used to learn W) are shown as filled
circles; block-missing subjects projected via available views are shown as
open triangles. Returns a ggplot object that can be further customized with
+ layers.
## S3 method for class 'woven' plot(x, labels = NULL, dims = c(1L, 2L), highlight_anchors = TRUE, ...)## S3 method for class 'woven' plot(x, labels = NULL, dims = c(1L, 2L), highlight_anchors = TRUE, ...)
x |
a woven object from [woven()] |
labels |
integer or factor of length n for coloring points. If NULL, all points are plotted in a single color. |
dims |
integer vector of length 2: which latent dimensions to plot (default c(1, 2)) |
highlight_anchors |
logical: distinguish anchors from projected subjects via point shape (default TRUE) |
... |
unused; present for S3 compatibility |
a ggplot object, invisibly. The plot is printed as a side effect.
set.seed(1) n <- 20; K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) Y <- rep(1:2, each = n / 2) miss <- matrix(FALSE, n, 2) miss[c(15, 17, 19), 1] <- TRUE miss[c(16, 18, 20), 2] <- TRUE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA anchor_idx <- which(rowSums(miss) == 0) fit <- woven(list(X1, X2), Y = Y, anchor_idx = anchor_idx, K = K) plot(fit, labels = Y)set.seed(1) n <- 20; K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) Y <- rep(1:2, each = n / 2) miss <- matrix(FALSE, n, 2) miss[c(15, 17, 19), 1] <- TRUE miss[c(16, 18, 20), 2] <- TRUE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA anchor_idx <- which(rowSums(miss) == 0) fit <- woven(list(X1, X2), Y = Y, anchor_idx = anchor_idx, K = K) plot(fit, labels = Y)
Print method for WOVEN fit
## S3 method for class 'woven' print(x, ...)## S3 method for class 'woven' print(x, ...)
x |
a woven object from [woven()] |
... |
further arguments (unused) |
Invisibly returns the woven object x.
set.seed(1) n <- 20 K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) Y <- rep(1:2, each = n / 2) miss <- matrix(FALSE, n, 2) miss[c(15, 17, 19), 1] <- TRUE miss[c(16, 18, 20), 2] <- TRUE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA anchor_idx <- which(rowSums(miss) == 0) fit <- woven(list(X1, X2), Y = Y, anchor_idx = anchor_idx, K = K) print(fit)set.seed(1) n <- 20 K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) Y <- rep(1:2, each = n / 2) miss <- matrix(FALSE, n, 2) miss[c(15, 17, 19), 1] <- TRUE miss[c(16, 18, 20), 2] <- TRUE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA anchor_idx <- which(rowSums(miss) == 0) fit <- woven(list(X1, X2), Y = Y, anchor_idx = anchor_idx, K = K) print(fit)
Prints a compact metrics table: silhouette, Davies-Bouldin, NMI, and ESS for the scored subjects. Requires class labels.
## S3 method for class 'woven' summary(object, labels = NULL, ...)## S3 method for class 'woven' summary(object, labels = NULL, ...)
object |
a woven object from [woven()] |
labels |
character, factor, or integer vector of length n. |
... |
unused |
Invisibly returns a named numeric vector of metrics.
data(woven_example) fit <- woven(woven_example$X_complete, Y = woven_example$Y, K = 3L) summary(fit, labels = woven_example$Y)data(woven_example) fit <- woven(woven_example$X_complete, Y = woven_example$Y, K = 3L) summary(fit, labels = woven_example$Y)
Learns a shared supervised latent space across V omics modalities, handling block-missing data via anchor-restricted alignment and Nystrm projection. Labels Y are required – WOVEN is a supervised method (cf. DIABLO).
woven( X_list, Y, anchor_idx = NULL, K = 5L, lambdas = 0.1, gamma_y = 1, k_nn = 10L, precomp = NULL, verbose = TRUE )woven( X_list, Y, anchor_idx = NULL, K = 5L, lambdas = 0.1, gamma_y = 1, k_nn = 10L, precomp = NULL, verbose = TRUE )
X_list |
list of V numeric matrices, each n x p_v. Subjects missing an entire modality should have that matrix row set to NA. |
Y |
integer or factor vector of length n – class labels for all subjects. Only anchor subjects' labels enter the supervised objective. |
anchor_idx |
integer vector – indices of fully-observed subjects (observed in all V modalities). Must have length >= K. If NULL (default), anchors are detected automatically as subjects with no block-missing modalities. |
K |
integer – number of latent dimensions (default 5) |
lambdas |
numeric scalar or length-V vector – Laplacian regularization strength per modality (default 0.1 for all) |
gamma_y |
numeric >= 0 – supervision strength. 0 = unsupervised CCA. Default 1.0 (equal weight to cross-modal alignment and label alignment). Tune via cross-validation on anchor set if labels are noisy. |
k_nn |
integer – k-nearest-neighbors for Laplacian graph (default 10).
Ignored when |
precomp |
optional output of [woven_precompute()] – pre-built Laplacians.
Pass this when calling |
verbose |
logical – print progress (default TRUE) |
For V=2, uses the closed-form supervised CCA solver (fast, exact). For V>=3, uses the ALS solver with label-kernel supervision.
object of class "woven" with:
n x K matrix of consensus latent scores for ALL n subjects (anchors and block-missing). The primary output for downstream analysis.
list of V projection matrices, each p_v x K
list of V anchor latent score matrices, each n_a x K
K supervised canonical correlations
indices of anchor (fully-observed) subjects
class label levels used during fitting
dimensions
hyperparameters
[woven_scores()], [woven_predict()], [woven_all_metrics()]
set.seed(1) n <- 60; p1 <- 20; p2 <- 15; K <- 2 Y <- rep(1:2, each = n / 2) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) # 30% block missingness; enforce >= 1 view per subject miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA # anchor_idx auto-detected from NA pattern -- no need to specify fit <- woven(list(X1, X2), Y = Y, K = K) dim(fit$Z) # 60 x 2 -- all subjects scoredset.seed(1) n <- 60; p1 <- 20; p2 <- 15; K <- 2 Y <- rep(1:2, each = n / 2) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) # 30% block missingness; enforce >= 1 view per subject miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA # anchor_idx auto-detected from NA pattern -- no need to specify fit <- woven(list(X1, X2), Y = Y, K = K) dim(fit$Z) # 60 x 2 -- all subjects scored
DB = (1/K) sum_i max_{j != i} (s_i + s_j) / d(c_i, c_j) where s_i = mean intra-cluster distance, d(c_i, c_j) = centroid distance.
woven_davies_bouldin(Z, labels)woven_davies_bouldin(Z, labels)
Z |
numeric matrix n x K |
labels |
integer or factor of length n |
scalar >= 0, lower is better
set.seed(1) Z <- matrix(rnorm(20 * 2), 20, 2) labels <- rep(1:2, each = 10) woven_davies_bouldin(Z, labels)set.seed(1) Z <- matrix(rnorm(20 * 2), 20, 2) labels <- rep(1:2, each = 10) woven_davies_bouldin(Z, labels)
Fits a linear model of a continuous outcome on a binary treatment indicator, separately within each subgroup defined by 'labels'. Compares estimated treatment effect to the known true effect (from simulation ground truth).
woven_effect_bias(Z, outcome, treatment, labels, true_effects)woven_effect_bias(Z, outcome, treatment, labels, true_effects)
Z |
numeric matrix n x K (latent scores; used as covariates) |
outcome |
numeric vector of length n (simulated continuous outcome) |
treatment |
integer/logical vector of length n (0/1 treatment indicator) |
labels |
integer or factor of length n (subgroup labels) |
true_effects |
named numeric vector, true treatment effect per subgroup level |
bias_g = |estimated_g - true_g| / |true_g| (relative) Returns mean bias across subgroups.
scalar >= 0, lower is better
set.seed(1) n <- 60 Z <- matrix(rnorm(n * 3), n, 3) outcome <- rnorm(n) treatment <- rep(0:1, n / 2) labels <- rep(1:2, each = n / 2) true_eff <- c(0.5, 1.0) woven_effect_bias(Z, outcome, treatment, labels, true_eff)set.seed(1) n <- 60 Z <- matrix(rnorm(n * 3), n, 3) outcome <- rnorm(n) treatment <- rep(0:1, n / 2) labels <- rep(1:2, each = n / 2) true_eff <- c(0.5, 1.0) woven_effect_bias(Z, outcome, treatment, labels, true_eff)
Effective sample size retention
woven_ess_retention(n_used, n_total)woven_ess_retention(n_used, n_total)
n_used |
integer, number of subjects with a latent score |
n_total |
integer, total subjects in dataset |
scalar in [0, 1], higher is better (DIABLO structurally caps at overlap fraction)
woven_ess_retention(n_used = 80, n_total = 100)woven_ess_retention(n_used = 80, n_total = 100)
A small simulated three-modality dataset (90 subjects) for illustrating package functions. Parameters and label structure are inspired by typical ADNI multi-omics studies (CN / MCI / AD), but all values are synthetic.
woven_examplewoven_example
A list with three components:
List of three matrices: RNA (90 x 25 genes), Methylation (90 x 20 CpGs), Proteomics (90 x 15 proteins). No missing values.
Same three matrices with ~33% of subjects missing one entire modality block (MCAR).
Character vector of 90 class labels: "CN", "MCI", "AD" (30 subjects each).
data(woven_example) # Complete data fit <- woven(woven_example$X_complete, Y = woven_example$Y, K = 3L) summary(fit, labels = woven_example$Y) # Block-missing data — WOVEN retains all 90 subjects fit_miss <- woven(woven_example$X_missing, Y = woven_example$Y, K = 3L) woven_metrics(fit_miss, woven_example$Y)data(woven_example) # Complete data fit <- woven(woven_example$X_complete, Y = woven_example$Y, K = 3L) summary(fit, labels = woven_example$Y) # Block-missing data — WOVEN retains all 90 subjects fit_miss <- woven(woven_example$X_missing, Y = woven_example$Y, K = 3L) woven_metrics(fit_miss, woven_example$Y)
Single eigendecomposition of a (V*n_a) x (V*n_a) block matrix. No iterations, no random restarts, no local optima. Unified solver for all V.
woven_mcca_dual( X_list, anchor_idx, Y, K = 5L, lambdas = 0.1, gamma_y = 1, k_nn = 10L, La_list_precomp = NULL, verbose = TRUE )woven_mcca_dual( X_list, anchor_idx, Y, K = 5L, lambdas = 0.1, gamma_y = 1, k_nn = 10L, La_list_precomp = NULL, verbose = TRUE )
X_list |
list of V matrices, each n x p_v (NA rows = block-missing) |
anchor_idx |
integer vector of fully-observed subject indices |
Y |
vector of length n, class labels (required) |
K |
integer, number of latent dimensions |
lambdas |
numeric scalar or length-V vector, Laplacian regularization |
gamma_y |
numeric >= 0, label supervision strength |
k_nn |
integer, k-NN for Laplacian (ignored if La_list_precomp supplied) |
La_list_precomp |
optional list of pre-extracted n_a x n_a anchor Laplacians |
verbose |
logical |
list with W_list, Za_list, Xa_list, singular_values, and metadata. Compatible with project_all() in benchmark_one_rep.R.
set.seed(1) n <- 20 K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) X3 <- matrix(rnorm(n * 3), n, 3) Y <- rep(1:2, each = n / 2) anchor_idx <- seq_len(14L) fit <- woven_mcca_dual(list(X1, X2, X3), anchor_idx = anchor_idx, Y = Y, K = K) length(fit$W_list)set.seed(1) n <- 20 K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) X3 <- matrix(rnorm(n * 3), n, 3) Y <- rep(1:2, each = n / 2) anchor_idx <- seq_len(14L) fit <- woven_mcca_dual(list(X1, X2, X3), anchor_idx = anchor_idx, Y = Y, K = K) length(fit$W_list)
Calls [woven_all_metrics()] using fit$Z and fit$n so you
do not need to extract them manually. Returns silhouette, Davies-Bouldin,
NMI, and ESS retention as a named numeric vector.
woven_metrics(fit, labels, ...)woven_metrics(fit, labels, ...)
fit |
woven object from [woven()] |
labels |
integer, factor, or character vector of length n with subgroup labels for all subjects (same Y passed to [woven()]) |
... |
additional arguments forwarded to [woven_all_metrics()] |
named numeric vector of metric values, printed as a tidy table
[woven_all_metrics()], [woven_silhouette()], [woven_nmi()]
set.seed(1) n <- 40; K <- 2L X1 <- matrix(rnorm(n * 8), n, 8) X2 <- matrix(rnorm(n * 6), n, 6) Y <- rep(1:2, each = n / 2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_metrics(fit, Y)set.seed(1) n <- 40; K <- 2L X1 <- matrix(rnorm(n * 8), n, 8) X2 <- matrix(rnorm(n * 6), n, 6) Y <- rep(1:2, each = n / 2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_metrics(fit, Y)
Uses k-means on Z to get cluster assignments, then computes NMI. k-means run 10 times to reduce initialization variance.
woven_nmi(Z, labels, n_cl = NULL, n_start = 10L)woven_nmi(Z, labels, n_cl = NULL, n_start = 10L)
Z |
numeric matrix n x K |
labels |
integer or factor of length n (true labels) |
n_cl |
integer, number of clusters (default = number of unique labels) |
n_start |
integer, k-means random starts |
scalar in [0, 1], higher is better
set.seed(1) Z <- matrix(rnorm(40 * 2), 40, 2) labels <- rep(1:2, each = 20) woven_nmi(Z, labels)set.seed(1) Z <- matrix(rnorm(40 * 2), 40, 2) labels <- rep(1:2, each = 20) woven_nmi(Z, labels)
For each held-out anchor subject, refits WOVEN without it, projects via direct W scoring, and computes ||Z_proj - Z_direct||. Quantifies how well the projection generalizes across anchor subsets.
woven_nystrom_error(fit, X_list, n_loo = NULL, sigma_proj = NULL)woven_nystrom_error(fit, X_list, n_loo = NULL, sigma_proj = NULL)
fit |
a woven object from [woven()] |
X_list |
list of complete (no block-missing) modality matrices, same structure as passed to [woven()] |
n_loo |
integer, number of anchors to hold out (default min(20, n_a)) |
sigma_proj |
unused, kept for compatibility |
scalar >= 0, lower is better (mean Frobenius error per anchor)
data(woven_example) fit <- woven(woven_example$X_complete, Y = woven_example$Y, K = 3L) woven_nystrom_error(fit, woven_example$X_complete, n_loo = 5L)data(woven_example) fit <- woven(woven_example$X_complete, Y = woven_example$Y, K = 3L) woven_nystrom_error(fit, woven_example$X_complete, n_loo = 5L)
For a given latent dimension, shows the top features by absolute loading
for each modality (or a selected subset), colored by loading sign.
Positive loadings are blue; negative loadings are red-orange.
Equivalent to DIABLO's plotLoadings().
woven_plot_loadings( fit, dim = 1L, n_top = 15L, feature_names = NULL, modality = NULL, main = NULL )woven_plot_loadings( fit, dim = 1L, n_top = 15L, feature_names = NULL, modality = NULL, main = NULL )
fit |
woven object from [woven()] |
dim |
integer: which latent dimension to plot (1..K, default 1) |
n_top |
integer: number of top features per modality (default 15) |
feature_names |
optional list of V character vectors (one per modality). If a plain character vector is passed for a single-modality call, it is used for that modality. If NULL, uses rownames of W or "Feature_j". |
modality |
integer or NULL: if specified, plot only that modality. If NULL (default), all V modalities are shown in faceted panels. |
main |
character: plot title. If NULL, a default is used. |
a ggplot object (printed automatically; add layers with +)
[woven_plot_vip()], [woven_plot_variance()]
set.seed(1) n <- 60; p1 <- 30; p2 <- 20; K <- 3 Y <- rep(1:3, each = n / 3) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_plot_loadings(fit, dim = 1L)set.seed(1) n <- 60; p1 <- 30; p2 <- 20; K <- 3 Y <- rep(1:3, each = n / 3) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_plot_loadings(fit, dim = 1L)
Bar chart of the proportion of variance explained per latent dimension (proportional to squared singular values), overlaid with a cumulative variance curve on a secondary axis. Use this to choose K and to show how much shared multi-omics signal is captured in the leading dimensions.
woven_plot_variance(fit, main = "Variance Explained")woven_plot_variance(fit, main = "Variance Explained")
fit |
woven object from [woven()] |
main |
character: plot title (default "Variance Explained") |
a ggplot object (printed automatically; add layers with +)
[woven_plot_vip()], [woven_plot_loadings()]
set.seed(1) n <- 60; p1 <- 30; p2 <- 20; K <- 4 Y <- rep(1:2, each = n / 2) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_plot_variance(fit)set.seed(1) n <- 60; p1 <- 30; p2 <- 20; K <- 4 Y <- rep(1:2, each = n / 2) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_plot_variance(fit)
Displays the top features by Variable Importance in Projection (VIP) score for one modality. VIP scores weight each feature's loading across all K latent dimensions by the variance explained per dimension, producing a single importance ranking analogous to DIABLO's contribution plot. A dashed reference line at VIP = 1 marks above-average importance.
woven_plot_vip( fit, modality = 1L, n_top = 20L, feature_names = NULL, main = NULL )woven_plot_vip( fit, modality = 1L, n_top = 20L, feature_names = NULL, main = NULL )
fit |
woven object from [woven()] |
modality |
integer: which modality to plot (1..V, default 1) |
n_top |
integer: number of top features to display (default 20) |
feature_names |
optional character vector of length p_v with feature labels. If NULL, uses rownames of W_list[[modality]] or "Feature_j". |
main |
character: plot title. If NULL, a default is generated. |
a ggplot object (printed automatically; add layers with +)
[woven_plot_loadings()], [woven_plot_variance()], [woven_vip()]
set.seed(1) n <- 60; p1 <- 30; p2 <- 20; K <- 3 Y <- rep(1:3, each = n / 3) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_plot_vip(fit, modality = 1L)set.seed(1) n <- 60; p1 <- 30; p2 <- 20; K <- 3 Y <- rep(1:3, each = n / 3) X1 <- matrix(rnorm(n * p1), n, p1) X2 <- matrix(rnorm(n * p2), n, p2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA fit <- woven(list(X1, X2), Y = Y, K = K) woven_plot_vip(fit, modality = 1L)
Builds k-NN RBF Laplacians for each modality from the observed data.
Pass the result to woven(..., precomp = precomp) to avoid
rebuilding the graph on every call – useful for hyperparameter search,
cross-validation, or sensitivity analysis.
woven_precompute(X_list, k_nn = 10L)woven_precompute(X_list, k_nn = 10L)
X_list |
list of V matrices (n x p_v). Block-missing rows (all NA) are automatically excluded from the k-NN graph. |
k_nn |
integer, number of nearest neighbours (default 10) |
list of V sparse Laplacian matrices, one per modality. Pass directly
to the precomp argument of [woven()].
[woven()]
set.seed(1) n <- 40; K <- 2L X1 <- matrix(rnorm(n * 8), n, 8) X2 <- matrix(rnorm(n * 6), n, 6) Y <- rep(1:2, each = n / 2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA; X2[miss[, 2], ] <- NA precomp <- woven_precompute(list(X1, X2), k_nn = 10L) fit <- woven(list(X1, X2), Y = Y, K = K, precomp = precomp)set.seed(1) n <- 40; K <- 2L X1 <- matrix(rnorm(n * 8), n, 8) X2 <- matrix(rnorm(n * 6), n, 6) Y <- rep(1:2, each = n / 2) miss <- matrix(runif(n * 2) < 0.3, n, 2) for (i in which(rowSums(miss) == 2)) miss[i, sample(2, 1)] <- FALSE X1[miss[, 1], ] <- NA; X2[miss[, 2], ] <- NA precomp <- woven_precompute(list(X1, X2), k_nn = 10L) fit <- woven(list(X1, X2), Y = Y, K = K, precomp = precomp)
Projects new subjects into the WOVEN latent space and returns soft class assignments using a nearest-centroid classifier in latent space. Works for complete subjects (direct projection) and block-missing subjects (Nystrm).
woven_predict(fit, X_list_new, method = "centroid", k_pred = 5L)woven_predict(fit, X_list_new, method = "centroid", k_pred = 5L)
fit |
woven object from [woven()] |
X_list_new |
list of V matrices for new subjects (n_new x p_v each). Block-missing subjects should have their modality rows set to NA. |
method |
"centroid" (default) – nearest centroid in latent space. "knn" – k-NN vote using anchor subjects as the reference set. |
k_pred |
integer – number of neighbors for knn method (default 5) |
data.frame with n_new rows: $predicted_class integer predicted class label $confidence probability of predicted class (0-1) One column per class level with soft probabilities
set.seed(1) n <- 40 K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) Y <- rep(1:2, each = n / 2) miss <- matrix(FALSE, n, 2) miss[c(31, 33, 35), 1] <- TRUE miss[c(32, 34, 36), 2] <- TRUE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA anchor_idx <- which(rowSums(miss) == 0) fit <- woven(list(X1, X2), Y = Y, anchor_idx = anchor_idx, K = K) pred <- woven_predict(fit, list(X1[1:5, ], X2[1:5, ])) pred$predicted_classset.seed(1) n <- 40 K <- 2L X1 <- matrix(rnorm(n * 5), n, 5) X2 <- matrix(rnorm(n * 4), n, 4) Y <- rep(1:2, each = n / 2) miss <- matrix(FALSE, n, 2) miss[c(31, 33, 35), 1] <- TRUE miss[c(32, 34, 36), 2] <- TRUE X1[miss[, 1], ] <- NA X2[miss[, 2], ] <- NA anchor_idx <- which(rowSums(miss) == 0) fit <- woven(list(X1, X2), Y = Y, anchor_idx = anchor_idx, K = K) pred <- woven_predict(fit, list(X1[1:5, ], X2[1:5, ])) pred$predicted_class
RV(X, Y) = trace(X X' Y Y') / sqrt(trace(X X' X X') * trace(Y Y' Y Y')) Measures similarity of two cross-product matrices; 1 = identical subspace.
woven_rv(Z, Z_true)woven_rv(Z, Z_true)
Z |
numeric matrix n x K (inferred latent scores) |
Z_true |
numeric matrix n x K_true (ground-truth factor scores from SUMO) |
scalar in [0, 1], higher is better
set.seed(1) Z <- matrix(rnorm(20 * 2), 20, 2) Z_true <- matrix(rnorm(20 * 3), 20, 3) woven_rv(Z, Z_true)set.seed(1) Z <- matrix(rnorm(20 * 2), 20, 2) Z_true <- matrix(rnorm(20 * 3), 20, 3) woven_rv(Z, Z_true)
Projects new subjects into the trained WOVEN latent space and returns an n_new x K score matrix. Uses direct linear projection (x %*% W_v) for each available modality, then averages across observed views.
For class predictions on new subjects, use woven_predict() instead.
woven_scores(fit, X_list_new)woven_scores(fit, X_list_new)
fit |
woven object from |
X_list_new |
list of V matrices (n_new x p_v). Set entire rows to NA for subjects missing that modality block. Every subject must have at least one non-missing view. |
Numeric matrix n_new x K of consensus latent scores. Subjects with no observed data in any view receive a row of NA.
woven_predict() for class predictions,
woven() for model fitting.
# minimal example data (n=20, 14 anchors, 6 partial) set.seed(1); n <- 20; K <- 2L X1 <- matrix(rnorm(n*5), n, 5); X2 <- matrix(rnorm(n*4), n, 4) Y <- rep(1:2, each = n/2) # Rows 15-20: alternate missing view 1 or view 2 (never both) miss <- matrix(FALSE, n, 2) miss[c(15,17,19), 1] <- TRUE # miss view 1 miss[c(16,18,20), 2] <- TRUE # miss view 2 X1[miss[,1],] <- NA; X2[miss[,2],] <- NA anchor_idx <- which(rowSums(miss)==0) fit <- woven(list(X1,X2),Y=Y,anchor_idx=anchor_idx,K=K) dim(woven_scores(fit, list(X1,X2)))# minimal example data (n=20, 14 anchors, 6 partial) set.seed(1); n <- 20; K <- 2L X1 <- matrix(rnorm(n*5), n, 5); X2 <- matrix(rnorm(n*4), n, 4) Y <- rep(1:2, each = n/2) # Rows 15-20: alternate missing view 1 or view 2 (never both) miss <- matrix(FALSE, n, 2) miss[c(15,17,19), 1] <- TRUE # miss view 1 miss[c(16,18,20), 2] <- TRUE # miss view 2 X1[miss[,1],] <- NA; X2[miss[,2],] <- NA anchor_idx <- which(rowSums(miss)==0) fit <- woven(list(X1,X2),Y=Y,anchor_idx=anchor_idx,K=K) dim(woven_scores(fit, list(X1,X2)))
Average silhouette width
woven_silhouette(Z, labels)woven_silhouette(Z, labels)
Z |
numeric matrix n x K (latent scores) |
labels |
integer or factor of length n (subgroup labels) |
scalar in [-1, 1], higher is better
set.seed(1) Z <- matrix(rnorm(20 * 2), 20, 2) labels <- rep(1:2, each = 10) woven_silhouette(Z, labels)set.seed(1) Z <- matrix(rnorm(20 * 2), 20, 2) labels <- rep(1:2, each = 10) woven_silhouette(Z, labels)