openproblems-bio · rcannood · Jul 31, 2026 · Jul 31, 2026 · Jul 31, 2026 · Jul 31, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -36,6 +36,14 @@
 
 * `novel_train`: Store only the metadata `novel_predict` needs in the model artifact, instead of a full copy of `train_mod2` (PR #47).
 
+* `lm`: Fit every column of mod2 in one `solve()` of the normal equations instead of calling `RcppArmadillo::fastLm()` once per column. Every column shares the same design matrix, so the old loop redid an `n x n_pcs` decomposition for each of the ~229k ATAC peaks. Predictions are unchanged; `RcppArmadillo` and `pbapply` are no longer needed (PR #58).
+
+* `novel_predict`, `simple_mlp_predict`, `ss_opm_predict`, `babel_predict`: Ask for `midgpu` rather than `gpu`. All four peak below 4 GB, so they fit the `de.NBI GPU T4 medium` flavour and no longer occupy a whole large node. The train steps stay on `gpu` (PR #58).
+
+* `knnr_py`, `knnr_r`, `cellmapper_linear`: Ask for `highmem` rather than `midmem`, and `lowcpu` rather than `midcpu` for the latter two. `knnr_py` peaks at 82 GB and `cellmapper_linear` at 68 GB against a 50 GB request, and both were OOM-killed on the larger datasets (PR #58).
+
+* `solution`, `zeros`: Ask for `lowmem` rather than `midmem` -- they use 0.5 GB and 11 GB of the 50 GB they asked for (PR #58).
+
 ## BUG FIXES
 
 * `guanlab_dengkw_pm`: Restore the consensus scheme of the original submission -- five reshuffles of the batches into two halves, ten kernel ridge models averaged. The port had replaced it with a single fixed two-way split for ADT pairs and leave-one-batch-out otherwise, so the result depended on the order the batches happened to come in. `--n_repeats` and `--seed` are now arguments; the unused `--distance_method` and `--n_pcs` are gone (PR #32).

diff --git a/common b/common
diff --git a/src/control_methods/solution/config.vsh.yaml b/src/control_methods/solution/config.vsh.yaml
@@ -13,4 +13,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [midtime, midmem, lowcpu]
+      label: [midtime, lowmem, lowcpu]
diff --git a/src/control_methods/zeros/config.vsh.yaml b/src/control_methods/zeros/config.vsh.yaml
@@ -13,4 +13,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [midtime, midmem, lowcpu]
+      label: [midtime, lowmem, lowcpu]
diff --git a/src/methods/babel/babel_predict/config.vsh.yaml b/src/methods/babel/babel_predict/config.vsh.yaml
@@ -25,4 +25,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [highmem, hightime, midcpu, gpu]
+      label: [highmem, hightime, midcpu, midgpu]
diff --git a/src/methods/cellmapper_linear/config.vsh.yaml b/src/methods/cellmapper_linear/config.vsh.yaml
@@ -71,4 +71,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [midtime,midmem,midcpu]
+      label: [midtime, highmem, lowcpu]
diff --git a/src/methods/knnr_py/config.vsh.yaml b/src/methods/knnr_py/config.vsh.yaml
@@ -33,4 +33,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [hightime, midmem, midcpu]
+      label: [hightime, highmem, midcpu]
diff --git a/src/methods/knnr_r/config.vsh.yaml b/src/methods/knnr_r/config.vsh.yaml
@@ -36,4 +36,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [hightime, midmem, midcpu]
+      label: [hightime, highmem, lowcpu]
diff --git a/src/methods/lm/config.vsh.yaml b/src/methods/lm/config.vsh.yaml
@@ -27,7 +27,7 @@ engines:
     image: openproblems/base_r:1
     setup:
       - type: r
-        cran: [ lmds, RcppArmadillo, pbapply]
+        cran: [ lmds ]
 runners:
   - type: executable
   - type: nextflow

diff --git a/src/methods/lm/script.R b/src/methods/lm/script.R
@@ -1,6 +1,5 @@
 cat("Loading dependencies\n")
 requireNamespace("anndata", quietly = TRUE)
-requireNamespace("pbapply", quietly = TRUE)
 library(Matrix, warn.conflicts = FALSE, quietly = TRUE)
 
 ## VIASH START
@@ -42,24 +41,14 @@ gc()
 cat("Reading mod2 files\n")
 X_mod2 <- anndata::read_h5ad(par$input_train_mod2)$layers[["normalized"]]
 
-cat("Predicting for each column in modality 2\n")
-preds <- pbapply::pblapply(
-  seq_len(ncol(X_mod2)),
-  function(i) {
-    y <- X_mod2[, i]
-    uy <- unique(y)
-    if (length(uy) > 1) {
-      fit <- RcppArmadillo::fastLm(dr_train, y)
-      # fit <- lm(y ~ ., dr_train)
-      stats::predict(fit, dr_test)
-    } else {
-      rep(uy, nrow(dr_test))
-    }
-  }
+cat("Predicting every column in modality 2 at once\n")
+coefficients <- solve(
+  crossprod(dr_train),
+  as.matrix(crossprod(dr_train, X_mod2))
 )
 
 cat("Creating outputs object\n")
-prediction <- Matrix::Matrix(do.call(cbind, preds), sparse = TRUE)
+prediction <- Matrix::Matrix(dr_test %*% coefficients, sparse = TRUE)
 rownames(prediction) <- rownames(dr_test)
 colnames(prediction) <- colnames(X_mod2)
 

diff --git a/src/methods/novel/novel_predict/config.vsh.yaml b/src/methods/novel/novel_predict/config.vsh.yaml
@@ -22,5 +22,5 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [highmem, hightime, midcpu, highsharedmem, gpu]
+      label: [highmem, hightime, midcpu, highsharedmem, midgpu]
 
diff --git a/src/methods/simple_mlp/simple_mlp_predict/config.vsh.yaml b/src/methods/simple_mlp/simple_mlp_predict/config.vsh.yaml
@@ -25,4 +25,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [highmem, hightime, midcpu, gpu, highsharedmem]
+      label: [highmem, hightime, midcpu, midgpu, highsharedmem]
diff --git a/src/methods/ss_opm/ss_opm_predict/config.vsh.yaml b/src/methods/ss_opm/ss_opm_predict/config.vsh.yaml
@@ -39,4 +39,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [highmem, hightime, midcpu, highsharedmem, gpu]
+      label: [highmem, hightime, midcpu, highsharedmem, midgpu]
+39 −8		nextflow_helpers/labels_denbi.config
+1 −0		scripts/render_results_report