diff --git a/combine_rows_lowmem.py b/combine_rows_lowmem.py
new file mode 100644
index 0000000..b50587c
--- /dev/null
+++ b/combine_rows_lowmem.py
@@ -0,0 +1,77 @@
+#!/usr/bin/env python3
+
+import argparse as ap
+import numpy as np
+import pandas as pd
+import pickle
+
+from tqdm import tqdm
+
+# Parse arguments
+parser = ap.ArgumentParser()
+parser.add_argument("files", nargs="+", type=str)
+parser.add_argument("-out", "--output", type=str)
+args = parser.parse_args()
+
+# Open first file to get targets
+print("Reading targets...", end="", flush=True)
+targets = np.loadtxt(args.files[0], usecols=1, dtype="U4")
+n_targets = len(targets)
+print("done")
+
+# Build DataFrame
+# The DF are used to ensure that distances and ligand similarities are inserted at the correct place
+# Initializing the DF with a numpy array is essential for speed at assignment
+print("Allocating DataFrame memory...", end="", flush=True)
+df_dist = pd.DataFrame(
+    index=targets, columns=targets, data=-1 * np.ones((n_targets, n_targets))
+)
+df_lsim = pd.DataFrame(
+    index=targets, columns=targets, data=-1 * np.ones((n_targets, n_targets))
+)
+print("done")
+
+print("Merging data...", flush=True)
+for fname in tqdm(args.files):
+    target = np.loadtxt(fname, usecols=0, dtype="U4")[0]
+    ctargets = np.loadtxt(fname, usecols=1, dtype="U4")
+    dist = np.loadtxt(fname, usecols=2)
+    lsim = np.loadtxt(fname, usecols=3)
+
+    # Populate distance matrix
+    if len(dist) == n_targets:
+        df_dist.loc[target, ctargets] = dist
+    else:
+        print("  Invalid number of distances for {target}")
+
+    # Populate ligand similarity matrix
+    if len(lsim) == n_targets:
+        df_lsim.loc[target, ctargets] = lsim
+    else:
+        print("  Invalid number of ligand similarities for {target}")
+
+dist = df_dist.values
+lsim = df_lsim.values
+
+# Check properties
+print("Checking matrix properties...", flush=True)
+ddist, dlsim = np.diagonal(dist), np.diagonal(lsim)
+assert int(round(np.sum(ddist))) == int(round(np.sum(ddist[ddist < 0])))
+assert int(round(np.sum(dlsim[dlsim >= 0]))) - int(round(np.sum(dlsim[dlsim < 0]))) == n_targets
+print("done")
+
+# Set NaNs for compatibility with original implementation
+dist[dist < 0] = np.nan
+lsim[lsim < 0] = np.nan
+
+print("Checking data...", flush=True)
+rows, cols = np.where(np.isnan(dist))  # Invalid distances
+for t1, t2 in zip(df_dist.index.values[rows], df_dist.columns.values[cols]):
+    print(f"  Missing distance for {t1} {t2}")
+rows, cols = np.where(np.isnan(lsim))  # Invalid ligand similarities
+for t1, t2 in zip(df_dist.index.values[rows], df_dist.columns.values[cols]):
+    print(f"  Missing ligand similarity for {t1} {t2}")
+
+print(f"Dumping pickle object {args.output}...", end="", flush=True)
+pickle.dump((dist, targets, lsim), open(f"{args.output}", "wb"), -1)
+print("done")