出身豪门也还是会坑--Azimuth也是Bug成堆

从事生物信息分析，研究方向越前沿，就要面临越多来自信息侧的问题。即使是文章发的非常好，原作者共享了代码，甚至写了现成软件工具，也不代表我们能轻松顺利的用这些现成的东西来复现或进行研究，混乱的环境设置只是一方面，更多时候，由于软件作者并不是专业的软件工程师，工具功能大致能用已经谢天谢地了，不能奢望这些软件没有一点毛病，更不能奢望有性能可言（除非开发时性能被来就是开发点）。即使这工具来自于有实力的大实验室，很多时候也不能幸免，比如… Azimuth。

Azimuth 介绍

Azimuth是由Satija实验室开发的单细胞数据注释工具，旨在简化 Seurat 中的 Label Transfer 流程，快速的对待分型细胞进行 Label Transfer。

问题描述

在当前最新的 0.5.0 版本中，在Jupyter Notebook 中运行 AzimuthReference 会提示 Error in ValidateAzimuthReference(object = object): Reference must contain an AzimuthData object in the tools slot.

谷歌一搜，发现该问题在2024年4月就已经被提出（issue #219）了，提出问题的用户已经给了解决方案，但这个Issue至今开放，且问题仍未被解决（上周又有用户反馈遇到同样的问题）。

解决思路

如前所述，Issue发起人zacharyrs已经提到了问题所在，在b1b6895这个提交中，代码作者使用sys.calls()来判断当前调用的函数名称，并根据这个名称来判断后续处理。

tool.name <- as.character(x = sys.calls()) # <-- 这里的sys.calls()会返回一个list，如果不是直接调用AzimuthReference，就会删除对象中的一些信息
tool.name <- lapply(
  X = strsplit(x = tool.name, split = "(", fixed = TRUE), 
  FUN = "[", 
  1
)[[1]]
if (tool.name != "AzimuthReference") {
  slot(object, name = "tools")["AzimuthReference"] <- slot(object, name = "tools")[tool.name]
  slot(object, name = "tools")[tool.name] <- NULL
}
object <- DietSeurat(object = object, 
                      counts = FALSE, 
                      assays = c("refAssay", assays), 
                      dimreducs = c("refDR", "refUMAP"))

然而实际上我们很多时候需要将AzimuthReference包到函数里使用，这时候，sys.calls() 返回的list中，第一项内容是最外侧的函数名称，这样就会导致Azimuth运行需要的数据被删除，下面的检查就会报错了。

对于我的情况来说，我也是第一次知道，notebook中所有的代码都是包在IRKernel的函数下的（感谢AI给了我排查方法），所以即使我笔记中直接运行AzimuthReference，也会发生将AzimuthReference放在函数中运行时才触发的错误。

处理也很简单，修改这里的判断，sys.calls()返回list后，去最后一项，再做正则取函数名就行：

1 2	call_list <- sys.calls() tool.name <- as.character(x = call_list[[length(call_list)]])

在我们的实际代码中，则可以加载Azimuth后，自行用修改后的函数去覆盖原来的AzimuthReference，这样后续代码就能运行了。

library(Azimuth)
AzimuthReference <- function (object, refUMAP = "umap", refDR = "spca", refAssay = "SCT", 
    dims = 1:50, k.param = 31, plotref = "umap", plot.metadata = NULL, 
    ori.index = NULL, colormap = NULL, assays = NULL, metadata = NULL, 
    reference.version = "0.0.0", verbose = FALSE) 
{
    if (!refUMAP %in% Reductions(object = object)) {
        stop("refUMAP (", refUMAP, ") not found in Seurat object provided")
    }
    if (is.null(x = Misc(object = object[[refUMAP]], slot = "model"))) {
        stop("refUMAP (", refUMAP, ") does not have the umap model info stored. ", 
            "Please rerun RunUMAP with return.model = TRUE.")
    }
    if (!refDR %in% Reductions(object = object)) {
        stop("refDR (", refDR, ") not found in Seurat object provided")
    }
    if (is.null(x = metadata)) {
        stop("Please specify at least one metadata field (for transfer and plotting).")
    }
    for (i in metadata) {
        if (!i %in% colnames(x = object[[]])) {
            warning(i, " not found in Seurat object metadata")
            next
        }
        if (!is.factor(x = object[[i, drop = TRUE]])) {
            warning(i, " is not a factor. Converting to factor with alphabetical ", 
                "levels.", call. = FALSE)
            object[[i, drop = TRUE]] <- factor(x = object[[i, 
                drop = TRUE]], levels = sort(x = unique(object[[i, 
                drop = TRUE]])))
        }
    }
    if (!refAssay %in% Assays(object = object)) {
        stop("Seurat object provided must have the SCT Assay stored.")
    }
    if (!inherits(x = object[[refAssay]], what = "SCTAssay")) {
        stop("refAssay (", refAssay, ") is not an SCTAssay.")
    }
    if (length(x = levels(x = object[[refAssay]])) != 1) {
        stop("refAssay (", refAssay, ") should contain a single SCT model.")
    }
    suppressWarnings(expr = object[["refUMAP"]] <- object[[refUMAP]])
    suppressWarnings(expr = object[["refDR"]] <- object[[refDR]])
    object <- FindNeighbors(object = object, reduction = "refDR", 
        dims = dims, graph.name = "refdr.annoy.neighbors", k.param = k.param, 
        nn.method = "annoy", annoy.metric = "cosine", cache.index = TRUE, 
        return.neighbor = TRUE, l2.norm = FALSE, verbose = verbose)
    if (verbose) {
        message("Computing pseudobulk averages")
    }
    features <- rownames(x = Loadings(object = object[["refDR"]]))
    plot.metadata <- plot.metadata %||% object[[metadata]]
    if (inherits(x = plotref, what = "DimReduc")) {
        plot.metadata <- plot.metadata[Cells(x = plotref), ]
    }
    ad <- CreateAzimuthData(object = object, plotref = plotref, 
        plot.metadata = plot.metadata, colormap = colormap, reference.version = reference.version)
    ori.index <- ori.index %||% match(Cells(x = object), Cells(x = object[["refUMAP"]]))
    object$ori.index <- ori.index
    DefaultAssay(object = object) <- refAssay
    object[[refAssay]] <- subset(x = object[[refAssay]], features = features)
    DefaultAssay(object = object[["refDR"]]) <- refAssay
    object <- DietSeurat(object = object, counts = FALSE, assays = c(refAssay, 
        assays), dimreducs = c("refDR", "refUMAP"))
    metadata <- c(metadata, "ori.index")
    for (i in colnames(x = object[[]])) {
        if (!i %in% metadata) {
            object[[i]] <- NULL
        }
    }
    sct.model <- slot(object = object[[refAssay]], name = "SCTModel.list")[[1]]
    object[["refAssay"]] <- as(object = suppressWarnings(Seurat:::CreateDummyAssay(assay = object[[refAssay]])), 
        Class = "SCTAssay")
    slot(object = object[["refAssay"]], name = "SCTModel.list") <- list(refmodel = sct.model)
    DefaultAssay(object = object) <- "refAssay"
    DefaultAssay(object = object[["refDR"]]) <- "refAssay"
    Tool(object = object) <- ad
    call_list <- sys.calls()                                           # use the last element of sys.call() to get the right function name
    tool.name <- as.character(x = call_list[[length(call_list)]])      # use the last element of sys.call() to get the right function name
    tool.name <- lapply(X = strsplit(x = tool.name, split = "(", 
        fixed = TRUE), FUN = "[", 1)[[1]]
    if (tool.name != "AzimuthReference") {
        slot(object, name = "tools")["AzimuthReference"] <- slot(object, 
            name = "tools")[tool.name]
        slot(object, name = "tools")[tool.name] <- NULL
    }
    object <- DietSeurat(object = object, counts = FALSE, assays = c("refAssay", 
        assays), dimreducs = c("refDR", "refUMAP"))
    ValidateAzimuthReference(object = object)
    return(object)
}