Package 'GenomeScaleEmbeddings'

Attach a houba file and return the bigmemory::big.matrix

Description

Attach a houba file and return the bigmemory::big.matrix

Usage

attachHoubaBigMatrix(houba_file)

Arguments

houba_file

Path to houba file (without .desc extension)

Value

bigmemory::big.matrix object

---

Copy remote parquet files into a local DuckDB database file using explicit URLs

Description

Copy remote parquet files into a local DuckDB database file using explicit URLs

Usage

CopyParquetToDuckDB(
  db_path = "local_embeddings.duckdb",
  urlList = DatasetParquetUrlList(),
  table_name = "embeddings",
  overwrite = FALSE
)

Arguments

db_path	Path to the local DuckDB database file
urlList	Character vector of parquet URLs
table_name	Name of the table to create in the DuckDB database
overwrite	Whether to overwrite the existing database

---

Compute correlation between PC scores and genomic position, per chromosome

Description

Compute correlation between PC scores and genomic position, per chromosome

Usage

correlatePCWithPosition(pc_scores, info_df, pc = 1, method = "spearman")

Arguments

pc_scores	Matrix of principal component scores (variants x PCs)
info_df	Data frame with variant info (must contain 'chrom' and 'pos')
pc	Which principal component to correlate (default: 1)
method	Correlation method (default: 'spearman')

Value

Named vector of correlation values per chromosome

---

List of the huggingface datasets for the paper "Incorporating LLM Embeddings for Variation Across the Human Genome"

Description

List of the huggingface datasets for the paper "Incorporating LLM Embeddings for Variation Across the Human Genome"

Usage

DatasetParquetUrlList()

Value

A character vector of dataset names

---

Quick summary for houba mmatrix

Description

Quick summary for houba mmatrix

Usage

embeddingSummary(embMat)

Arguments

embMat

houba mmatrix

Value

List with dim, colMeans, rowMeans

---

Get PCA scores from houbaPCA result

Description

Get PCA scores from houbaPCA result

Usage

getPcaScores(houbaPCA_res)

Arguments

houbaPCA_res

List returned by houbaPCA (with 'pca' and 'houbaM')

Value

Matrix of principal component scores (variants x PCs)

---

PCA using bigPCAcpp on houba mmatrix or bigmemory::big.matrix

Description

PCA using bigPCAcpp on houba mmatrix or bigmemory::big.matrix

Usage

houbaPCA(
  embMat = "local_embeddings.houba",
  center = TRUE,
  scale = TRUE,
  ncomp = 15
)

Arguments

embMat	houba mmatrix path or bigmemory::big.matrix
center	logical, whether to center columns
scale	logical, whether to scale columns
ncomp	number of principal components to compute

Value

List with PCA result object and houbaM big.matrix

---

Quick summary for houba info mmatrix

Description

Quick summary for houba info mmatrix

Usage

infoSummary(infoMat)

Arguments

infoMat

houba mmatrix

Value

List with dim, unique chroms, and example rsids

---

Iterate over embeddings as matrix batches from a local DuckDB file

Description

Iterate over embeddings as matrix batches from a local DuckDB file

Usage

IterateEmbeddingsMatrixBatches(
  chunk_size = 1e+05,
  db_path = "local_embeddings.duckdb",
  table_name = "embeddings"
)

Arguments

chunk_size	Number of rows per batch
db_path	Path to the local DuckDB database file
table_name	Name of the table in the DuckDB database

Value

A list of embedding batches

---

Open remote parquet files as a DuckDB VIEW and return as tibble (minimal, http(s) only)

Description

Open remote parquet files as a DuckDB VIEW and return as tibble (minimal, http(s) only)

Usage

OpenRemoteParquetView(
  urlList = DatasetParquetUrlList(),
  view_name = "embeddings",
  db_path = tempfile(fileext = ".duckdb"),
  unify_schemas = FALSE
)

Arguments

urlList	Character vector of parquet URLs
view_name	Name of the DuckDB view to create
db_path	Path to DuckDB database file (default: temporary file, null for in-memory)
unify_schemas	Whether to unify schemas across files

Value

dplyr tibble referencing the DuckDB view

---

Plot PCA dimensions using ggplot2, colored by annotation

Description

Plot PCA dimensions using ggplot2, colored by annotation

Usage

plotPcaDims(pc_scores, info_df, annotation_col = "chrom", dim1 = 1, dim2 = 2)

Arguments

pc_scores	Matrix of principal component scores (variants x PCs)
info_df	Data frame with variant info (must contain annotation column)
annotation_col	Name of column in info_df to color by (e.g. 'gwas')
dim1	First PC dimension to plot (default: 1)
dim2	Second PC dimension to plot (default: 2)

---

Plot spatial correlation between PC scores and genomic position, faceted by chromosome

Description

Plot spatial correlation between PC scores and genomic position, faceted by chromosome

Usage

plotPCSpatialCorrelation(pc_scores, info_df, pc = 1)

Arguments

pc_scores	Matrix of principal component scores (variants x PCs)
info_df	Data frame with variant info (must contain 'chrom' and 'pos')
pc	Which principal component to plot (default: 1)

---

Write embeddings to houba mmatrix and return info as data.frame from a local DuckDB file

Description

Write embeddings to houba mmatrix and return info as data.frame from a local DuckDB file

Usage

writeEmbeddingsHoubaFromDuckDB(
  dbPath = "local_embeddings.duckdb",
  tableName = "embeddings",
  embeddingCol = "embedding",
  batchSize = 1e+05,
  embeddingDim = 3072,
  embeddingFile = gsub("\\.duckdb$", ".houba", dbPath),
  overwrite = FALSE
)

Arguments

dbPath	Path to the local DuckDB database file
tableName	Name of the table in the DuckDB database
embeddingCol	Name of the embeddings column
batchSize	Number of rows per batch
embeddingDim	Dimension of each embedding vector
embeddingFile	Path for houba mmatrix file
overwrite	Whether to overwrite existing houba file

Value

An object of class 'HoubaEmbeddings' with mmatrix, info data.frame, and houba file path

---