R/plot.R

In NCBI-Hackathons/clustifyR: Classifier for Single-cell RNA-seq Using Cell Clusters

#' Plot a tSNE or umap colored by feature.
#'
#' @param data input data
#' @param x x variable
#' @param y y variable
#' @param feature feature to color by
#' @param legend_name legend name to display, defaults to no name
#' @param c_cols character vector of colors to build color gradient
#' for continuous values, defaults to [`clustifyr::pretty_palette`]
#' @param d_cols character vector of colors for discrete values.
#' defaults to RColorBrewer paired palette
#' @param pt_size point size
#' @param alpha_col whether to refer to data column for alpha values
#' @param group_col group by another column instead of feature,
#'  useful for labels
#' @param scale_limits defaults to min = 0, max = max(data$x),
#' otherwise a two-element numeric vector indicating min and max to plot
#' @param do_label whether to label each cluster at median center
#' @param do_legend whether to draw legend
#' @param do_repel whether to use ggrepel on labels
#' @return ggplot object, cells projected by dr, colored by feature
#' @examples
#' plot_dims(
#'     pbmc_meta,
#'     feature = "classified"
#' )
#' @export
plot_dims <- function(data,
    x = "UMAP_1",
    y = "UMAP_2",
    feature = NULL,
    legend_name = "",
    c_cols = pretty_palette2,
    d_cols = NULL,
    pt_size = 0.25,
    alpha_col = NULL,
    group_col = NULL,
    scale_limits = NULL,
    do_label = FALSE,
    do_legend = TRUE,
    do_repel = TRUE) {
    # add backticks to allow special characters in column names

    # If feature is not provided return unlabeled plot
    if (is.null(feature)) {
        p <- ggplot(data, aes(.data[[x]], .data[[y]])) +
            geom_point(size = pt_size) +
            cowplot::theme_cowplot()

        if (!is.null(d_cols)) {
            p <- p +
                scale_color_manual(values = d_cols)
        }

        return(p)
    }

    # Retrieve features from data
    feature_data <- data[[feature]]
    n_features <- length(unique(feature_data))

    feature_types <- c(
        "character",
        "logical",
        "factor"
    )

    # If there are too many features, add more colors for pretty_palette2
    if (identical(c_cols, pretty_palette2) &
        (n_features > length(pretty_palette2)) &
        (typeof(feature_data) %in% feature_types)) {
        c_cols <- pretty_palette_ramp_d(n_features)
        d_cols <- pretty_palette_ramp_d(n_features)
    }

    # sort data to avoid plotting null values over colors
    data <- dplyr::arrange(data, !!dplyr::sym(feature))

    if (!is.null(alpha_col)) {
        p <- ggplot(data, aes(.data[[x]], .data[[y]])) +
            geom_point(aes(
                color = .data[[feature]],
                alpha = .data[[alpha_col]]
            ), # backticks protect special character gene names
            size = pt_size
            ) +
            scale_alpha_continuous(range = c(0, 1))
    } else {
        p <- ggplot(data, aes(.data[[x]], .data[[y]])) +
            geom_point(aes(color = .data[[feature]]),
            size = pt_size
            )
    }

    # discrete values
    if (!is.numeric(feature_data)) {
        # use colors provided
        if (!is.null(d_cols)) {
            p <- p +
                scale_color_manual(
                    values = d_cols,
                    name = legend_name
                )
        } else {
            p <- p +
                scale_color_brewer(
                    palette = "Paired",
                    name = legend_name
                )
        }

        # continuous values
    } else {
        if (is.null(scale_limits)) {
            scale_limits <- c(
                ifelse(min(feature_data) < 0, min(feature_data), 0),
                max(feature_data)
            )
        }

        p <- p +
            scale_color_gradientn(
                colors = c_cols,
                name = legend_name,
                limits = scale_limits
            )
    }

    if (do_label) {
        if (is.null(group_col)) {
            centers <- dplyr::group_by(data, !!sym(feature))
        } else {
            centers <- dplyr::group_by(data, !!sym(feature), !!sym(group_col))
        }

        if (!(is.null(alpha_col))) {
            centers <-
                dplyr::summarize(centers,
                    t1 = median(!!dplyr::sym(x)),
                    t2 = median(!!dplyr::sym(y)),
                    a = median(!!dplyr::sym(alpha_col))
                )
            centers <- dplyr::ungroup(centers)

            if (!(is.null(group_col))) {
                centers <- dplyr::select(centers, -!!sym(group_col))
            }
        } else {
            centers <-
                dplyr::summarize(centers,
                    t1 = median(!!dplyr::sym(x)),
                    t2 = median(!!dplyr::sym(y)),
                    a = 1
                )
            centers <- dplyr::ungroup(centers)

            if (!(is.null(group_col))) {
                centers <- dplyr::select(centers, -!!sym(group_col))
            }
        }

        if (do_repel) {
            alldata <- dplyr::select(
                data,
                !!dplyr::sym(feature),
                !!dplyr::sym(x),
                !!dplyr::sym(y)
            )
            alldata[[1]] <- ""
            alldata$a <- 0
            colnames(alldata) <- colnames(centers)
            alldata <- rbind(alldata, centers)
            p <- p +
                geom_point(
                    data = alldata,
                    mapping = aes(
                        x = !!dplyr::sym("t1"),
                        y = !!dplyr::sym("t2")
                    ),
                    size = 3,
                    alpha = 0
                ) +
                ggrepel::geom_text_repel(
                    data = alldata,
                    mapping = aes(
                        x = !!dplyr::sym("t1"),
                        y = !!dplyr::sym("t2"),
                        alpha = !!dplyr::sym("a"),
                        label = .data[[feature]]
                    ),
                    point.padding = 0.5,
                    box.padding = 0.5,
                    max.iter = 50000
                )
        } else {
            p <- p +
                geom_text(
                    data = centers,
                    mapping = aes(
                        x = !!dplyr::sym("t1"),
                        y = !!dplyr::sym("t2"),
                        label = centers[[feature]]
                    ),
                    alpha = centers[["a"]]
                )
        }
    }

    p <- p +
        cowplot::theme_cowplot()

    if (!do_legend) {
        p <- p +
            theme(legend.position = "none")
    }

    p
}

#' Color palette for plotting continous variables
#' @return vector of colors
pretty_palette <- rev(scales::brewer_pal(palette = "RdGy")(6))

#' Color palette for plotting continous variables, starting at gray
#' @return vector of colors
pretty_palette2 <- scales::brewer_pal(palette = "Reds")(9)

#' black and white palette for plotting continous variables
#' @return vector of colors
not_pretty_palette <- scales::brewer_pal(palette = "Greys")(9)

#' Expanded color palette ramp for plotting discrete variables
#' @param n number of colors to use
#' @return color ramp
pretty_palette_ramp_d <-
    grDevices::colorRampPalette(scales::brewer_pal(palette = "Paired")(12))

#' Plot similarity measures on a tSNE or umap
#'
#' @param cor_mat input similarity matrix
#' @param metadata input metadata with per cell tsne or
#'   umap coordinates and cluster ids
#' @param data_to_plot colname of data to plot, defaults to all
#' @param cluster_col colname of clustering data in metadata, defaults
#'  to rownames of the metadata if not supplied.
#' @param x metadata column name with 1st axis dimension.
#' defaults to "UMAP_1".
#' @param y metadata column name with 2nd axis dimension.
#' defaults to "UMAP_2".
#' @param scale_legends if TRUE scale all legends to maximum values in entire
#' correlation matrix. if FALSE scale legends to maximum for each plot. A
#' two-element numeric vector can also be passed to supply
#' custom values i.e. c(0, 1)
#' @param ... passed to plot_dims
#' @return list of ggplot objects, cells projected by dr,
#' colored by cor values
#' @examples
#' res <- clustify(
#'     input = pbmc_matrix_small,
#'     metadata = pbmc_meta,
#'     ref_mat = cbmc_ref,
#'     query_genes = pbmc_vargenes,
#'     cluster_col = "classified"
#' )
#'
#' plot_cor(
#'     cor_mat = res,
#'     metadata = pbmc_meta,
#'     data_to_plot = colnames(res)[1:2],
#'     cluster_col = "classified",
#'     x = "UMAP_1",
#'     y = "UMAP_2"
#' )
#' @export
plot_cor <- function(cor_mat,
    metadata,
    data_to_plot = colnames(cor_mat),
    cluster_col = NULL,
    x = "UMAP_1",
    y = "UMAP_2",
    scale_legends = FALSE,
    ...) {
    cor_matrix <- cor_mat
    if (!any(data_to_plot %in% colnames(cor_matrix))) {
        stop("cluster ids not shared between metadata and correlation matrix",
            call. = FALSE
        )
    }

    if (is.null(cluster_col)) {
        cluster_col <- "rownames"
        metadata <- tibble::rownames_to_column(metadata, cluster_col)
    }

    cor_df <- as.data.frame(cor_matrix)
    cor_df <- tibble::rownames_to_column(cor_df, cluster_col)
    cor_df_long <- tidyr::gather(
        cor_df,
        "ref_cluster",
        "expr", -dplyr::matches(cluster_col)
    )

    # If cluster_col is factor, convert to character
    if (is.factor(metadata[, cluster_col])) {
        metadata[, cluster_col] <- as.character(metadata[, cluster_col])
    }

    # checks matrix rownames,
    # 2 branches for cluster number (avg) or cell bar code (each cell)
    if (cor_df[[cluster_col]][1] %in% metadata[[cluster_col]]) {
        plt_data <- dplyr::left_join(cor_df_long,
            metadata,
            by = cluster_col,
            relationship = "many-to-many"
        )
    } else {
        plt_data <- dplyr::left_join(cor_df_long,
            metadata,
            by = structure(names = cluster_col, "rn")
        )
    }

    # determine scaling method, either same for all plots,
    # or per plot (default)
    if (is.logical(scale_legends) && scale_legends) {
        scale_limits <- c(
            ifelse(min(plt_data$expr) < 0,
                min(plt_data$expr),
                0
            ),
            max(max(plt_data$expr))
        )
    } else if (is.logical(scale_legends) && !scale_legends) {
        scale_limits <- NULL
    } else {
        scale_limits <- scale_legends
    }

    plts <- vector("list", length(data_to_plot))

    for (i in seq_along(data_to_plot)) {
        tmp_data <- dplyr::filter(
            plt_data,
            !!dplyr::sym("ref_cluster") == data_to_plot[i]
        )

        plts[[i]] <- plot_dims(
            data = tmp_data,
            x = x,
            y = y,
            feature = "expr",
            legend_name = data_to_plot[i],
            scale_limits = scale_limits,
            ...
        )
    }

    plts
}

#' Plot gene expression on to tSNE or umap
#'
#' @param expr_mat input single cell matrix
#' @param metadata data.frame with tSNE or umap coordinates
#' @param genes gene(s) to color tSNE or umap
#' @param cell_col column name in metadata containing cell ids, defaults
#' to rownames if not supplied
#' @param ... additional arguments passed to `[clustifyr::plot_dims()]`
#' @return list of ggplot object, cells projected by dr,
#' colored by gene expression
#' @examples
#' genes <- c(
#'     "RP11-314N13.3",
#'     "ARF4"
#' )
#'
#' plot_gene(
#'     expr_mat = pbmc_matrix_small,
#'     metadata = tibble::rownames_to_column(pbmc_meta, "rn"),
#'     genes = genes,
#'     cell_col = "rn"
#' )
#' @export
plot_gene <- function(expr_mat,
    metadata,
    genes,
    cell_col = NULL,
    ...) {
    genes_to_plot <- genes[genes %in% rownames(expr_mat)]
    genes_missing <- setdiff(genes, genes_to_plot)

    if (length(genes_missing) != 0) {
        message(
            "the following genes were not present in the input matrix ",
            paste(genes_missing, collapse = ",")
        )
    }

    if (length(genes_to_plot) == 0) {
        stop("no genes present to plot", call. = FALSE)
    }
    expr_dat <- t(as.matrix(expr_mat[genes_to_plot, , drop = FALSE]))
    expr_dat <-
        tibble::rownames_to_column(as.data.frame(expr_dat), "cell")

    if (is.null(cell_col)) {
        mdata <- tibble::rownames_to_column(metadata, "cell")
        cell_col <- "cell"
    } else {
        mdata <- metadata
    }

    if (!cell_col %in% colnames(mdata)) {
        stop("please supply a cell_col that is present in metadata",
            call. = FALSE
        )
    }

    plt_dat <- dplyr::left_join(expr_dat, mdata,
        by = c("cell" = cell_col)
    )

    lapply(
        genes_to_plot,
        function(gene) {
            plot_dims(plt_dat,
                feature = gene,
                legend_name = gene,
                ...
            )
        }
    )
}

#' Plot called clusters on a tSNE or umap, for each reference cluster given
#'
#' @param cor_mat input similarity matrix
#' @param metadata input metadata with tsne or
#'  umap coordinates and cluster ids
#' @param data_to_plot colname of data to plot, defaults to all
#' @param ... passed to plot_dims
#' @return list of ggplot object, cells projected by dr,
#' colored by cell type classification
plot_call <- function(cor_mat,
    metadata,
    data_to_plot = colnames(cor_mat),
    ...) {
    cor_matrix <- cor_mat
    df_temp <-
        as.data.frame(cor_matrix - matrixStats::rowMaxs(as.matrix(cor_matrix)))
    df_temp[df_temp == 0] <- "1"
    df_temp[df_temp != "1"] <- "0"
    plot_cor(
        df_temp,
        metadata,
        data_to_plot,
        ...
    )
}

#' Plot best calls for each cluster on a tSNE or umap
#'
#' @param cor_mat input similarity matrix
#' @param metadata input metadata with tsne or
#'  umap coordinates and cluster ids
#' @param cluster_col metadata column, can be cluster or cellid
#' @param collapse_to_cluster if a column name is provided,
#' takes the most frequent call of entire cluster to color in plot
#' @param threshold minimum correlation coefficent cutoff for calling clusters
#' @param x x variable
#' @param y y variable
#' @param plot_r whether to include second plot of cor eff for best call
#' @param per_cell whether the cor_mat was generate per cell or per cluster
#' @param ... passed to plot_dims
#' @return ggplot object, cells projected by dr,
#'  colored by cell type classification
#' @examples
#' res <- clustify(
#'     input = pbmc_matrix_small,
#'     metadata = pbmc_meta,
#'     ref_mat = cbmc_ref,
#'     query_genes = pbmc_vargenes,
#'     cluster_col = "classified"
#' )
#'
#' plot_best_call(
#'     cor_mat = res,
#'     metadata = pbmc_meta,
#'     cluster_col = "classified"
#' )
#' @export
plot_best_call <- function(cor_mat,
    metadata,
    cluster_col = "cluster",
    collapse_to_cluster = FALSE,
    threshold = 0,
    x = "UMAP_1",
    y = "UMAP_2",
    plot_r = FALSE,
    per_cell = FALSE,
    ...) {
    cor_matrix <- cor_mat
    col_meta <- colnames(metadata)
    if ("type" %in% col_meta | "type2" %in% col_meta) {
        warning('metadata column name clash of "type"/"type2"')
        return()
    }
    df_temp <- cor_to_call(
        cor_matrix,
        metadata = metadata,
        cluster_col = cluster_col,
        threshold = threshold
    )

    df_temp_full <- call_to_metadata(
        df_temp,
        metadata = metadata,
        cluster_col = cluster_col,
        per_cell = per_cell
    )

    if (collapse_to_cluster != FALSE) {
        df_temp_full <- collapse_to_cluster(df_temp_full,
            metadata,
            collapse_to_cluster,
            threshold = threshold
        )
    }

    g <- plot_dims(df_temp_full,
        feature = "type",
        x = x,
        y = y,
        ...
    )

    if (plot_r) {
        l <- list()
        l[[1]] <- g
        l[[2]] <- plot_dims(df_temp_full,
            feature = "r",
            x = x,
            y = y,
            ...
        )
    } else {
        l <- g
    }

    l
}

#' Plot similarity measures on heatmap
#'
#' @param cor_mat input similarity matrix
#' @param metadata input metadata with per cell tsne
#' or umap cooordinates and cluster ids
#' @param cluster_col colname of clustering data in metadata,
#'  defaults to rownames of the metadata if not supplied.
#' @param col color ramp to use
#' @param legend_title legend title to pass to Heatmap
#' @param ... passed to Heatmap
#' @return complexheatmap object
#' @examples
#' res <- clustify(
#'     input = pbmc_matrix_small,
#'     metadata = pbmc_meta,
#'     ref_mat = cbmc_ref,
#'     query_genes = pbmc_vargenes,
#'     cluster_col = "classified",
#'     per_cell = FALSE
#' )
#'
#' plot_cor_heatmap(res)
#' @export
plot_cor_heatmap <- function(cor_mat,
    metadata = NULL,
    cluster_col = NULL,
    col = not_pretty_palette,
    legend_title = NULL,
    ...) {
    cor_matrix <- cor_mat
    ComplexHeatmap::Heatmap(
        cor_matrix,
        col = col,
        heatmap_legend_param = list(title = legend_title),
        ...
    )
}