FindNeighbors not work in Seurat 4.0.0 but work well in Seurat 3.2.3.

[RuzhangZhao]

I just updated Seurat to the newest version, but found FindNeighbors does not work for my current case. Therefore, I tested several datasets, including the example data from Seurat website. The FindNeighbors function still not work. I also tried to export the pca matrix out and directly fed into FindNeighbors function. It will not work in as.matrix version or as.data.frame version.

By the way, I use mac under macOS 10.15.4, R studio with R version 4.0.4, Seurat 4.0.0. The code example below use the provided data from Seurat website pbmc3k guided tutorial.

If I follow the below codes, the output will be

Computing nearest neighbor graph
Error: node stack overflow
Error during wrapup: node stack overflow
Error: no more error handlers available (recursive errors?); invoking 'abort' restart

If I directly feed matrix into FindNeighbors, (previously works well for me) the output is the same as above.
If I directly feed data.frame into FindNeighbors,(previously works well for me) it will report

Computing nearest neighbors
Error in a$addItem(ii - 1, data[ii, ]) : 
  Not compatible with requested type: [type=list; target=double].

# insert reproducible example here
library(dplyr)
library(Seurat)
library(patchwork)
expr<- Read10X(data.dir = "filtered_gene_bc_matrices/hg19/")
expr<-CreateSeuratObject(expr)
expr<- NormalizeData(expr)
expr<-FindVariableFeatures(expr)
expr<-ScaleData(expr)
expr<-RunPCA(expr, features = VariableFeatures(object = expr))
expr <- FindNeighbors(expr, dims = 1:10)

[RuzhangZhao]

I think I found the reason. Once I load package Rfast, the error message will appear. There may be some conflict between newly updated Seurat and Rfast. Hope this can help you to reproduce the error message.

[NMikolajewicz]

I've started receiving this error message recently. Any progress in resolving it? In my case it doesn't seem to be related to Rfast.

[RuzhangZhao]

Hi Nicholas. Previously, I had a list of packages. I excluded each one of them from the list by order and tried to use the Find Neighbor function. Finally, my case is that if I excluded Rfast, it will work. Would you mind sharing your case? I think that might help the team to resolve the problem.

[NMikolajewicz]

Here is my script:

`
packages2load <- c("Seurat", "scater", "scMiko", "reticulate", "destiny", "foreach", "doParallel", "clusterExperiment",
"plyr", "dplyr", "tidyr", "reshape2", "tidyverse", "RColorBrewer", "ggplot2", "gridExtra",
"DT", "flexdashboard", "ggpmisc", "ggpmisc", "ggExtra", "grid", "ggrepel", "scales", "ggformula",
"tidymodels", "viridis", "future", "diffusr", "markovchain", "Tempora", "s2a", "glue")

load packages

invisible({lapply(packages2load, library, character.only = TRUE)})

process expresion matrix

so.wgcna <- Seurat::CreateSeuratObject(counts = mat.data)
so.wgcna <- Seurat::ScaleData(so.wgcna)
so.wgcna <- Seurat::RunPCA(so.wgcna, features = rownames(so.wgcna), weight.by.var = F)
so.wgcna <- Seurat::FindNeighbors(so.wgcna, k.param = 5, features = rownames(so.wgcna), reduction = "pca")
so.wgcna <- Seurat::RunUMAP(so.wgcna, dim = 1:50, n.neighbors = 10, min.dist = 0.001, metric = "cosine")`

[NMikolajewicz]

Hi again. I managed to resolve the problem by changing nn.method from "annoy" to "rann". Seemed to be related to the "annoy" method in my case.

[RuzhangZhao]

Hi Nicholas. Thanks so much! That also works for me.