scClassify:::getResByWeights issue

[yulijia]

I'm trying to use the PBMC model to predict cell type labels for an in-house dataset:

pred_res <- predict_scClassifyJoint(exprsMat_test = exprsMat,
                               trainRes = pbmc_model, 
                               algorithm = "WKNN",
                               features = c("limma"),
                               similarity = c("pearson", "spearman"),
                               prob_threshold = 0.7,
                               verbose = TRUE)

Since I’m using both "pearson" and "spearman" as similarities, does that mean the predictions are ensembled across these methods?

I noticed that in the scClassify:::calEnsembleRes function (link here), a weight vector of length 60 is generated. Then weight[keep] is passed to scClassify:::getResByWeights, but it still contains all 60 values.

Then the following code breaks if res only includes two values

 > res
                                pearson_WKNN_limma                                spearman_WKNN_limma 
                                     "CD4+ T cell" "CD4+ T cell_Cytotoxic T cell_Natural killer cell"

##break
    mat <- vapply(seq_len(length(resType)), function(i) {
      ifelse(res %in% resType[i], 1, 0) * weight
    }, numeric(length(res)))

To fix it, I think it should be:

mat <- vapply(seq_len(length(resType)), function(i) {
  as.numeric(res == resType[i]) * weight
}, numeric(length(weight)))

But even after fixing that, I ran into another issue when assigning row and column names:

colnames(mat) <- resType
rownames(mat) <- names(res)

This fails because mat has 60 rows (from weight), but names(res) only has 2 values: "pearson_WKNN_limma" and "spearman_WKNN_limma".

So I'm a bit confused — is this a bug in how getResByWeights() expects the input, or am I misunderstanding how the ensembling is supposed to work?