From 3fb1a6110de3eaf1a444f49d92b090ee75341e09 Mon Sep 17 00:00:00 2001
From: Neil Thomas <neil@evolutionaryscale.ai>
Date: Fri, 4 Apr 2025 10:55:40 -0700
Subject: [PATCH] 3.2.0 - fix residue constants (#231)

---
 esm/utils/residue_constants.py       | 26 ++++++++++++++++++++++++++
 esm/utils/structure/protein_chain.py |  3 +--
 2 files changed, 27 insertions(+), 2 deletions(-)

diff --git a/esm/utils/residue_constants.py b/esm/utils/residue_constants.py
index 86ea82e..b1af539 100644
--- a/esm/utils/residue_constants.py
+++ b/esm/utils/residue_constants.py
@@ -79,3 +79,29 @@ restype_1to3 = {
     "Y": "TYR",
     "V": "VAL",
 }
+
+# Approximate Volumes of amino acids in cubic angstroms.
+# https://www.imgt.org/IMGTeducation/Aide-memoire/_UK/aminoacids/abbreviation.html
+amino_acid_volumes = {
+    "A": 88.6,  # Alanine
+    "R": 173.4,  # Arginine
+    "N": 114.1,  # Asparagine
+    "D": 111.1,  # Aspartic acid
+    "C": 108.5,  # Cysteine
+    "Q": 143.8,  # Glutamine
+    "E": 138.4,  # Glutamic acid
+    "G": 60.1,  # Glycine
+    "H": 153.2,  # Histidine
+    "I": 166.7,  # Isoleucine
+    "L": 166.7,  # Leucine
+    "K": 168.6,  # Lysine
+    "M": 162.9,  # Methionine
+    "F": 189.9,  # Phenylalanine
+    "P": 112.7,  # Proline
+    "S": 89.0,  # Serine
+    "T": 116.1,  # Threonine
+    "W": 227.8,  # Tryptophan
+    "Y": 193.6,  # Tyrosine
+    "V": 140.0,  # Valine
+    "X": 88.6,  # Unknown, use Alanine as approximation
+}
diff --git a/esm/utils/structure/protein_chain.py b/esm/utils/structure/protein_chain.py
index 8a0ddf0..0b198e2 100644
--- a/esm/utils/structure/protein_chain.py
+++ b/esm/utils/structure/protein_chain.py
@@ -21,7 +21,6 @@ from scipy.spatial import ConvexHull
 from scipy.spatial.distance import pdist, squareform
 from torch import Tensor
 
-from evolutionaryscale import residue_constants
 from esm.utils import residue_constants as RC
 from esm.utils.constants import esm3 as C
 from esm.utils.misc import slice_python_object_as_numpy
@@ -322,7 +321,7 @@ class ProteinChain:
         sequence = [aa for aa, m in zip(self.sequence, mask) if m]
         A, _ = self._mvee(points, tol=1e-3)
         mvee_volume = (4 * np.pi) / (3 * np.sqrt(np.linalg.det(A)))
-        volume = sum(residue_constants.amino_acid_volumes[x] for x in sequence)
+        volume = sum(RC.amino_acid_volumes[x] for x in sequence)
         ratio = volume / mvee_volume
 
         # The paper says you must compare the ellipsoidal profile with T, a measurement of