add fraction matching based on only the light / na labels

R/utils_fractions.R

@@ -265,16 +265,30 @@
 #' @keywords internal
 .removeOverlappingFeatures = function(input) {
     Fraction = NULL
-    
+
     if (data.table::uniqueN(input$Fraction) > 1) {
         measurement_count = input[
+            (IsotopeLabelType == "L" | is.na(IsotopeLabelType)) &
             !is.na(Intensity) & Intensity > 0,
             .(n_obs = uniqueN(Run)),
             by = .(feature, Fraction)
         ]
+        # Features absent from measurement_count have no L or NA observations.
+        # Fall back to H observations so they are not silently dropped.
+        h_only_features = setdiff(unique(input$feature), unique(measurement_count$feature))
+        if (length(h_only_features) > 0) {
+            h_count = input[
+                feature %in% h_only_features &
+                    IsotopeLabelType == "H" &
+                    !is.na(Intensity) & Intensity > 0,
+                .(n_obs = uniqueN(Run)),
+                by = .(feature, Fraction)
+            ]
+            measurement_count = rbind(measurement_count, h_count)
+        }
         measurement_count[, is_max := n_obs == max(n_obs), by = "feature"]
         max_fractions = measurement_count[(is_max)]
-        
+
         fraction_map = .resolveFractionTies(input, max_fractions)
         input = input[fraction_map, on = .(feature, Fraction), nomatch = 0]
     }
@@ -293,14 +307,28 @@
 #' @keywords internal
 .resolveFractionTies = function(input, max_fractions) {
     tie_features = max_fractions[, .(n_ties = .N), by = "feature"][n_ties > 1, feature]
-    
+
     if (length(tie_features) > 0) {
         tied_fractions = max_fractions[feature %in% tie_features, .(feature, Fraction)]
-        avg_abundance = input[
-            tied_fractions, on = .(feature, Fraction), nomatch = 0
-        ][!is.na(Intensity) & Intensity > 0,
-          .(mean_abundance = mean(Intensity)),
-          by = .(feature, Fraction)]
+        tied_input = input[tied_fractions, on = .(feature, Fraction), nomatch = 0]
+        avg_abundance = tied_input[
+            (IsotopeLabelType == "L" | is.na(IsotopeLabelType)) &
+                !is.na(Intensity) & Intensity > 0,
+            .(mean_abundance = mean(Intensity)),
+            by = .(feature, Fraction)
+        ]
+        # H-only tied features have no L or NA rows; fall back to H for tie-breaking.
+        h_only_tied = setdiff(tie_features, unique(avg_abundance$feature))
+        if (length(h_only_tied) > 0) {
+            h_avg = tied_input[
+                feature %in% h_only_tied &
+                    IsotopeLabelType == "H" &
+                    !is.na(Intensity) & Intensity > 0,
+                .(mean_abundance = mean(Intensity)),
+                by = .(feature, Fraction)
+            ]
+            avg_abundance = rbind(avg_abundance, h_avg)
+        }
         best_tied = avg_abundance[, .SD[which.max(mean_abundance)], by = "feature"]
         best_simple = max_fractions[
             !feature %in% tie_features,

inst/tinytest/test_fractions.R

@@ -65,6 +65,7 @@ fractionated = data.table::data.table(
     feature = rep(c("A", "B"), each = 6),
     Fraction = rep(rep(c(1, 2), each = 3), times = 2),
     Run = 1:12,
+    IsotopeLabelType = "L",
     Intensity = c(NA, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2)
 )
 ### More observations win
@@ -87,6 +88,7 @@ fractionated_third = data.table::data.table(
     feature = rep("A", 9),
     Fraction = c(rep(1, 3), rep(2, 3), rep(3, 3)),
     Run = 1:9,
+    IsotopeLabelType = "L",
     Intensity = c(1, 1, 1,    # Fraction 1: 3 obs, mean = 1 (ties for max n_obs)
                   2, 2, 2,    # Fraction 2: 3 obs, mean = 2 (ties for max n_obs)
                   10, NA, NA) # Fraction 3: 1 obs, mean = 10 (loses on n_obs but mean would win w/o fix)
@@ -95,6 +97,73 @@ expect_equal(
     unique(MSstatsConvert:::.removeOverlappingFeatures(fractionated_third)$Fraction),
     2
 )
+### L/H isotope label types: fraction selection uses only L obs, output keeps both L and H;
+### NA-only features: fraction selection uses NA obs
+# Feature "A" (L+H): F1 has 1 L obs (run 1) + 4 H obs (runs 1-4) = 4 unique runs total
+#                    F2 has 3 L obs (runs 5-7) + 1 H obs (run 5)  = 3 unique runs total
+#   → If counting all unique runs: F1 wins (4 > 3)
+#   → If counting L unique runs only: F2 wins (3 > 1)
+#   → Expected: F2 selected, confirming H obs do not influence fraction selection
+#   → Both L and H rows from F2 are returned
+# Feature "B" (NA only): F1 has 2 NA obs (runs 8-9), F2 has 4 NA obs (runs 10-13)
+#   → NA obs: F1=2, F2=4 → F2 wins; all 4 F2 rows are kept
+fractionated_lh = data.table::data.table(
+    feature = c(rep("A", 9), rep("B", 6)),
+    Fraction = c(rep(1, 5), rep(2, 4),
+                 rep(1, 2), rep(2, 4)),
+    Run = c(1, 1, 2, 3, 4,    # A F1: run 1 paired L+H, runs 2-4 H only
+            5, 5, 6, 7,        # A F2: run 5 paired L+H, runs 6-7 L only
+            8, 9, 10, 11, 12, 13),
+    IsotopeLabelType = c("L", "H", "H", "H", "H",   # A F1: 1 L, 4 H
+                         "L", "H", "L", "L",          # A F2: 3 L, 1 H
+                         rep(NA_character_, 6)),
+    Intensity = rep(1, 15)
+)
+### Feature A: F1 has more total obs but fewer L obs — F2 must win to confirm L-only logic
+expect_equal(
+    unique(MSstatsConvert:::.removeOverlappingFeatures(fractionated_lh[feature == "A"])$Fraction),
+    2
+)
+### Both L and H rows are returned for the winning fraction
+expect_true(
+    all(c("L", "H") %in% MSstatsConvert:::.removeOverlappingFeatures(fractionated_lh[feature == "A"])$IsotopeLabelType)
+)
+### Feature B (NA only) selected via NA obs count
+expect_equal(
+    unique(MSstatsConvert:::.removeOverlappingFeatures(fractionated_lh[feature == "B"])$Fraction),
+    2
+)
+### Full dataset: A keeps F2 (L+H rows), B keeps F2 (NA rows)
+expect_identical(
+    MSstatsConvert:::.removeOverlappingFeatures(data.table::copy(fractionated_lh)),
+    fractionated_lh[(feature == "A" & Fraction == 2) | (feature == "B" & Fraction == 2)]
+)
+### H-only feature: falls back to H obs for fraction selection (not silently dropped)
+# F1 has 1 H obs, F2 has 3 H obs → F2 wins
+fractionated_h_only = data.table::data.table(
+    feature = rep("A", 4),
+    Fraction = c(1, 2, 2, 2),
+    Run = 1:4,
+    IsotopeLabelType = "H",
+    Intensity = 1
+)
+expect_equal(
+    unique(MSstatsConvert:::.removeOverlappingFeatures(fractionated_h_only)$Fraction),
+    2
+)
+### H-only feature with tied obs count: higher mean H intensity breaks the tie
+# F1: 2 obs, mean intensity 1; F2: 2 obs, mean intensity 3 → F2 wins
+fractionated_h_tied = data.table::data.table(
+    feature = rep("A", 4),
+    Fraction = rep(c(1, 2), each = 2),
+    Run = 1:4,
+    IsotopeLabelType = "H",
+    Intensity = c(1, 1, 3, 3)
+)
+expect_equal(
+    unique(MSstatsConvert:::.removeOverlappingFeatures(fractionated_h_tied)$Fraction),
+    2
+)
 fractionated_tmt = fractionated = data.table::data.table(
     feature = rep(c("A", "B"), each = 6),
     Fraction = rep(rep(c(1, 2), each = 3), times = 2),