> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/satijalab/seurat-wrappers/llms.txt
> Use this file to discover all available pages before exploring further.

# RunOptimizeALS()

> Run LIGER integrative NMF via optimizeALS on a Seurat object

`RunOptimizeALS()` wraps the LIGER `optimizeALS` algorithm to perform integrative non-negative matrix factorization (iNMF) on a merged Seurat object. Each dataset's cells share a common factor matrix W while having dataset-specific matrices H and V.

## Syntax

```r theme={null}
RunOptimizeALS(
  object,
  k,
  assay = NULL,
  split.by = "orig.ident",
  lambda = 5,
  thresh = 1e-6,
  max.iters = 30,
  reduction.name = "iNMF_raw",
  reduction.key = "riNMF_",
  nrep = 1,
  H.init = NULL,
  W.init = NULL,
  V.init = NULL,
  rand.seed = 1,
  print.obj = FALSE,
  ...
)
```

## Parameters

<ParamField path="object" type="Seurat" required>
  A merged Seurat object containing cells from multiple datasets.
</ParamField>

<ParamField path="k" type="integer" required>
  Number of factors (latent dimensions) to compute.
</ParamField>

<ParamField path="assay" type="character" default="NULL">
  Assay to use. Defaults to the active default assay.
</ParamField>

<ParamField path="split.by" type="character" default="orig.ident">
  Metadata column used to split cells into per-dataset subsets.
</ParamField>

<ParamField path="lambda" type="numeric" default="5">
  Regularization parameter. Larger values increase dataset-specific penalty.
</ParamField>

<ParamField path="thresh" type="numeric" default="1e-6">
  Convergence threshold for the ALS objective.
</ParamField>

<ParamField path="max.iters" type="integer" default="30">
  Maximum number of ALS iterations.
</ParamField>

<ParamField path="reduction.name" type="character" default="iNMF_raw">
  Name for the resulting DimReduc object.
</ParamField>

<ParamField path="reduction.key" type="character" default="riNMF_">
  Prefix for the iNMF embedding column names.
</ParamField>

<ParamField path="nrep" type="integer" default="1">
  Number of restarts. The best factorization is kept.
</ParamField>

<ParamField path="H.init" type="matrix" default="NULL">
  Initial value for the H (cell factor) matrices.
</ParamField>

<ParamField path="W.init" type="matrix" default="NULL">
  Initial value for the W (shared gene factor) matrix.
</ParamField>

<ParamField path="V.init" type="matrix" default="NULL">
  Initial value for the V (dataset-specific gene factor) matrices.
</ParamField>

<ParamField path="rand.seed" type="integer" default="1">
  Random seed for reproducibility.
</ParamField>

<ParamField path="print.obj" type="logical" default="FALSE">
  Print the objective value at each iteration.
</ParamField>

## Returns

A Seurat object with:

* A DimReduc under `reduction.name` containing iNMF cell embeddings and feature loadings
* Per-dataset feature loading matrices stored in the `Tool` slot (accessible via `Tool(object)`)

## Examples

```r theme={null}
library(SeuratWrappers)
install.packages("rliger")

# Preprocess: normalize and scale (without centering)
object <- NormalizeData(object)
object <- FindVariableFeatures(object)
object <- ScaleData(object, split.by = "orig.ident", do.center = FALSE)

# Run iNMF factorization
object <- RunOptimizeALS(object, k = 20, split.by = "orig.ident")

# Quantile normalize
object <- RunQuantileNorm(object, split.by = "orig.ident")

# Cluster and visualize
object <- FindNeighbors(object, reduction = "iNMF", dims = 1:ncol(Embeddings(object, "iNMF")))
object <- FindClusters(object)
object <- RunUMAP(object, reduction = "iNMF", dims = 1:ncol(Embeddings(object, "iNMF")))
```

## See Also

* [LIGER method guide](/methods/liger)