push new db or demo version

dump.rdb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edb6dac953afd60f97f39ca1a77dfb3cc323e573cbfd6b6ed768d6d637613898
-size 2454945
+oid sha256:d61b5bfb2340158a7febe75d096dbab4a1ab5a041d1f094cd18b34063fab3d2b
+size 89804397

src/app.py

@@ -44,7 +44,7 @@ class App:
     def _draw_molecule_grid(cls, similar: list[SimilarMolecule]) -> np.ndarray:
         mols = [Chem.MolFromSmiles(m["smiles"]) for m in similar]
         legends = [
-            f"{cls._truncated_attribute(m, 'name')}\n{m['category']}\n"
+            f"{cls._truncated_attribute(m, 'name')}\n{m['properties']}\n"
             f"{cls._truncated_attribute(m, 'smiles')}\n{m['score']:.2E}"
             for m in similar
         ]
@@ -57,7 +57,7 @@ class App:
         img = Draw.MolsToGridImage(
             mols,
             legends=legends,
-            molsPerRow=2,
+            molsPerRow=3,
             subImgSize=(250, 250),
             drawOptions=draw_options,
         )
@@ -67,7 +67,9 @@ class App:
     def _display_sample_molecules(mols: pd.DataFrame):
         for _, row in mols.iterrows():
             with gr.Group():
-                gr.Textbox(value=row["smiles"], label=f"{row['name']} ({row['category']})", interactive=False, scale=3)
+                gr.Textbox(
+                    value=row["smiles"], label=f"{row['name']} ({row['properties']})", interactive=False, scale=3
+                )
                 sample_btn = gr.Button(
                     f"Load {row['name']}",
                     scale=1,
@@ -81,7 +83,7 @@ class App:
 
     @staticmethod
     def clear_all():
-        return "", [], [], None, "Cleared - Draw a new molecule or enter SMILES"
+        return "", "", [], [], None, "Cleared - Draw a new molecule or enter SMILES"
 
     def handle_search(self, smiles: str, embed_dim: int):
         if not smiles.strip():
@@ -119,7 +121,7 @@ class App:
             """)
             gr.HTML(
                 """
-            The Redis database indexes a curated subset of molecules from <a href="https://isomerdesign.com/pihkal/home">Isomer Design</a>
+            The Redis database indexes <a href="https://isomerdesign.com/pihkal/home">Isomer Design</a> molecular library.
             <a href="https://creativecommons.org/licenses/by-nc-sa/4.0/">
                 <img src="https://mirrors.creativecommons.org/presskit/buttons/80x15/svg/by-nc-sa.svg" alt="License: CC BY-NC-SA 4.0"
              style="display:inline; height:15px; vertical-align:middle; margin-left:4px;"/>
@@ -141,6 +143,13 @@ class App:
                         elem_id="smiles_input",
                     )
 
+                    canonical_smiles_output = gr.Textbox(
+                        label="Canonical SMILES",
+                        placeholder="Canonical representation will appear here",
+                        interactive=False,
+                        elem_id="canonical_smiles_output",
+                    )
+
                     embedding_dimension = gr.Dropdown(
                         choices=SUPPORTED_EMBEDDING_DIMENSIONS,
                         value=EMBEDDING_DIMENSION,
@@ -193,6 +202,14 @@ class App:
                     with gr.Column(scale=1):
                         self._display_sample_molecules(SAMPLE_SMILES[2::3])
 
+            # Update canonical SMILES when input changes
+            smiles_input.change(
+                fn=self.embedding_service.get_canonical_smiles,
+                inputs=[smiles_input],
+                outputs=[canonical_smiles_output],
+                api_name="get_canonical_smiles",
+            )
+
             search_btn.click(
                 fn=self.handle_search,
                 inputs=[smiles_input, embedding_dimension],
@@ -211,6 +228,7 @@ class App:
                 js="window.clearJSME",
                 outputs=[
                     smiles_input,
+                    canonical_smiles_output,
                     embedding_output,
                     similar_molecules_output,
                     molecule_image,

src/constants.py

@@ -1,25 +1,25 @@
 # Model config
 MODEL_NAME = "Derify/ChemMRL-alpha"
-SUPPORTED_EMBEDDING_DIMENSIONS = [1024, 768, 512, 256, 128, 64, 32, 16, 8, 4, 2]
+SUPPORTED_EMBEDDING_DIMENSIONS = [1024, 768, 512, 256, 128, 64, 32, 16]
 EMBEDDING_DIMENSION = max(SUPPORTED_EMBEDDING_DIMENSIONS)
 USE_HALF_PRECISION = True
 
 # HNSW index parameters
-HNSW_K = 6
+HNSW_K = 9
 HNSW_PARAMETERS = {
     # Embedding vector dtype
     "TYPE": "FLOAT16" if USE_HALF_PRECISION else "FLOAT32",
     # Embedding vectors are normalized so COSINE and IP are equivalent
     "DISTANCE_METRIC": "IP",
     # Defines the initial capacity of the vector index. It helps in pre-allocating space for the index.
-    "INITIAL_CAP": 440,
+    "INITIAL_CAP": 15400,
     # Max number of outgoing edges (connections) for each node in a graph layer.
-    "M": 256,
+    "M": 32,
     # Max number of connected neighbors to consider during graph building.
     # Higher values increase accuracy, but also increase index build time.
-    "EF_CONSTRUCTION": 4096,
+    "EF_CONSTRUCTION": 512,
     # Max top candidates during KNN search. Higher values increase accuracy, but also increase search latency.
-    "EF_RUNTIME": 6,
+    "EF_RUNTIME": HNSW_K,
 }
 
 # Gradio launch parameters

src/data.py

@@ -4,10 +4,8 @@ import pandas as pd
 
 __data_dir = os.path.join(os.path.dirname(__file__), "data")
 
-__dataset_smiles_file = os.path.join(__data_dir, "dataset_smiles.csv")
 __sample_smiles_file = os.path.join(__data_dir, "sample_smiles.csv")
-__isomer_design_subset_file = os.path.join(__data_dir, "isomer_design_subset.csv")
+__isomer_design_dataset_file = os.path.join(__data_dir, "isomer_design_dataset.csv")
 
-DATASET_SMILES = pd.read_csv(__dataset_smiles_file)
 SAMPLE_SMILES = pd.read_csv(__sample_smiles_file)
-ISOMER_DESIGN_SUBSET = pd.read_csv(__isomer_design_subset_file)
+ISOMER_DESIGN_DATASET = pd.read_csv(__isomer_design_dataset_file, sep="\t")

src/data/sample_smiles.csv

@@ -1,4 +1,4 @@
-smiles,name,category
+smiles,name,properties
 CCO,Ethanol,Alcohol
 CC(=O)O,Acetic acid,Carboxylic acid
 c1ccccc1,Benzene,Aromatic hydrocarbon

src/service.py

@@ -1,5 +1,6 @@
 import logging
 import os
+import time
 from typing import TypedDict
 
 import numpy as np
@@ -21,7 +22,7 @@ from constants import (
     SUPPORTED_EMBEDDING_DIMENSIONS,
     USE_HALF_PRECISION,
 )
-from data import DATASET_SMILES, ISOMER_DESIGN_SUBSET
+from data import ISOMER_DESIGN_DATASET
 
 
 def setup_logger(clear_handler=False):
@@ -40,7 +41,7 @@ logger = setup_logger(clear_handler=True)
 class SimilarMolecule(TypedDict):
     smiles: str
     name: str
-    category: str
+    properties: str
     score: float
 
 
@@ -84,15 +85,24 @@ class MolecularEmbeddingService:
                 password=redis_password,
                 decode_responses=True,
             )
-            redis_client.ping()
-            return redis_client
         except Exception as e:
             logger.error(f"Failed to connect to Redis: {e}")
             raise
 
+        while True:
+            try:
+                redis_client.ping()
+                break
+            except redis.exceptions.BusyLoadingError:
+                time_out = 5
+                logger.warning(f"Redis is loading the dataset in memory. Retrying in {time_out} seconds...")
+                time.sleep(time_out)
+
+        return redis_client
+
     def _initialize_datastore(self):
         self.__create_hnsw_index()
-        self.__populate_sample_data(ISOMER_DESIGN_SUBSET)
+        self.__populate_sample_data(ISOMER_DESIGN_DATASET)
 
     def __create_hnsw_index(self):
         """Create HNSW index for molecular embeddings"""
@@ -195,14 +205,14 @@ class MolecularEmbeddingService:
             query = (
                 Query(f"*=>[KNN {k} @{self.embedding_field_name(embed_dim)} $vec AS score]")
                 .sort_by("score")
-                .return_fields("smiles", "name", "category", "score")
+                .return_fields("smiles", "name", "properties", "score")
                 .dialect(2)
             )
 
             results = self.redis_client.ft(self.index_name).search(query, query_params={"vec": query_vector})
 
             neighbors: list[SimilarMolecule] = [
-                {"smiles": doc.smiles, "name": doc.name, "category": doc.category, "score": float(doc.score)}
+                {"smiles": doc.smiles, "name": doc.name, "properties": doc.properties, "score": float(doc.score)}
                 for doc in results.docs
             ]
 
@@ -212,6 +222,16 @@ class MolecularEmbeddingService:
             logger.error(f"Failed to find similar molecules: {e}")
             return []
 
+    def get_canonical_smiles(self, smiles: str) -> str:
+        """Convert SMILES to canonical SMILES representation"""
+        if not smiles or smiles.strip() == "":
+            return ""
+
+        canonical = MorganFingerprinter.canonicalize_smiles(smiles.strip())
+        if canonical:
+            return canonical
+        return smiles.strip()
+
     @staticmethod
     def embedding_field_name(dim: int) -> str:
         return f"embedding_{dim}"

src/static/main.js

@@ -270,7 +270,7 @@ window.setJSMESmiles = function (smiles) {
 window.clearJSME = function () {
   jsmeApplet?.reset();
   updateGradioTextbox("");
-  return ["", [], [], "Cleared - Draw a new molecule or enter SMILES"];
+  return ["", "", [], [], "Cleared - Draw a new molecule or enter SMILES"];
 };
 
 // ============================================================================

src/static/main.min.js

@@ -1 +1 @@
-function initializeJSME(){try{jsmeApplet=new JSApplet.JSME("jsme_container",getJsmeContainerWidthPx(),"450px",{options:"NOcanonize,rButton,zoom,zoomgui,newLook,star,multipart,polarnitro,NOexportInChI,NOexportInChIkey,NOsearchInChIkey,NOexportSVG,NOpaste"}),jsmeApplet.setCallBack("AfterStructureModified",handleJSMEStructureChange),jsmeApplet.setMenuScale(getJsmeGuiScale()),jsmeApplet.setUserInterfaceBackgroundColor("#adadad"),jsmeApplet.readGenericMolecularInput("CCO"),lastTextboxValue="CCO",setupTextboxEventListeners(),window.addEventListener("resize",handleResize)}catch(e){throw e}}function handleJSMEStructureChange(){try{updateGradioTextbox(jsmeApplet.smiles())}catch(e){}}function getJsmeGuiScale(){const e=getJsmeContainerWidthNumber();if(null==e||e<=0)return 1;let t;return t=e>460?1.3:e>420?1.1:e>370?1.05:e>300?.88:2,t}function getJsmeContainerWidthPx(){const e=getJsmeContainerWidthNumber();return null==e||e<=0?"100%":`${e}px`}function getJsmeContainerWidthNumber(){const e=document.getElementById("jsme_container");return e?.parentNode?.offsetWidth}function updateGradioTextbox(e){try{const t=document.querySelector(SMILES_INPUT_SELECTOR);if(t?.value===e)return;t.value=e,lastTextboxValue=e,GRADIO_CHANGE_EVENTS.forEach((e=>{const n=new Event(e,{bubbles:!0,cancelable:!0});t.dispatchEvent(n)}))}catch(e){}}function updateJSMEFromTextbox(e){try{""!==e?.trim()?jsmeApplet?.readGenericMolecularInput(e.trim()):jsmeApplet?.reset(),lastTextboxValue=e}catch(e){}}function setupTextboxEventListeners(){const e=document.querySelector(SMILES_INPUT_SELECTOR);e&&(e.addEventListener("input",handleTextboxChange),e.addEventListener("change",handleTextboxChange),e.addEventListener("paste",handleTextboxPaste),e.addEventListener("keyup",handleTextboxChange))}function handleTextboxChange(e){e.target.value!==lastTextboxValue&&updateJSMEFromTextbox(e.target.value)}function handleTextboxPaste(e){setTimeout((()=>{updateJSMEFromTextbox(e.target.value)}),50)}function handleResize(){try{jsmeApplet?.setMenuScale(getJsmeGuiScale()),jsmeApplet?.setWidth(getJsmeContainerWidthPx())}catch(e){}}function initializeWhenReady(){"undefined"!=typeof JSApplet&&JSApplet.JSME?initializeJSME():setTimeout(initializeWhenReady,2e3)}function startInitialization(){"loading"===document.readyState?document.addEventListener("DOMContentLoaded",(()=>{setTimeout(initializeWhenReady,2e3)})):setTimeout(initializeWhenReady,2e3)}let jsmeApplet=null,lastTextboxValue="";const DEFAULT_SMILES="CCO",CONTAINER_HEIGHT="450px",SMILES_INPUT_SELECTOR="#smiles_input textarea, #smiles_input input",PASTE_DELAY=50,INIT_RETRY_DELAY=2e3,GRADIO_CHANGE_EVENTS=["input","change","keyup"];window.setJSMESmiles=function(e){return updateJSMEFromTextbox(e),updateGradioTextbox(e),e},window.clearJSME=function(){return jsmeApplet?.reset(),updateGradioTextbox(""),["",[],[],"Cleared - Draw a new molecule or enter SMILES"]},startInitialization();
+function initializeJSME(){try{jsmeApplet=new JSApplet.JSME("jsme_container",getJsmeContainerWidthPx(),"450px",{options:"NOcanonize,rButton,zoom,zoomgui,newLook,star,multipart,polarnitro,NOexportInChI,NOexportInChIkey,NOsearchInChIkey,NOexportSVG,NOpaste"}),jsmeApplet.setCallBack("AfterStructureModified",handleJSMEStructureChange),jsmeApplet.setMenuScale(getJsmeGuiScale()),jsmeApplet.setUserInterfaceBackgroundColor("#adadad"),jsmeApplet.readGenericMolecularInput("CCO"),lastTextboxValue="CCO",setupTextboxEventListeners(),window.addEventListener("resize",handleResize)}catch(e){throw e}}function handleJSMEStructureChange(){try{updateGradioTextbox(jsmeApplet.smiles())}catch(e){}}function getJsmeGuiScale(){const e=getJsmeContainerWidthNumber();if(null==e||e<=0)return 1;let t;return t=e>460?1.3:e>420?1.1:e>370?1.05:e>300?.88:2,t}function getJsmeContainerWidthPx(){const e=getJsmeContainerWidthNumber();return null==e||e<=0?"100%":`${e}px`}function getJsmeContainerWidthNumber(){const e=document.getElementById("jsme_container");return e?.parentNode?.offsetWidth}function updateGradioTextbox(e){try{const t=document.querySelector(SMILES_INPUT_SELECTOR);if(t?.value===e)return;t.value=e,lastTextboxValue=e,GRADIO_CHANGE_EVENTS.forEach((e=>{const n=new Event(e,{bubbles:!0,cancelable:!0});t.dispatchEvent(n)}))}catch(e){}}function updateJSMEFromTextbox(e){try{""!==e?.trim()?jsmeApplet?.readGenericMolecularInput(e.trim()):jsmeApplet?.reset(),lastTextboxValue=e}catch(e){}}function setupTextboxEventListeners(){const e=document.querySelector(SMILES_INPUT_SELECTOR);e&&(e.addEventListener("input",handleTextboxChange),e.addEventListener("change",handleTextboxChange),e.addEventListener("paste",handleTextboxPaste),e.addEventListener("keyup",handleTextboxChange))}function handleTextboxChange(e){e.target.value!==lastTextboxValue&&updateJSMEFromTextbox(e.target.value)}function handleTextboxPaste(e){setTimeout((()=>{updateJSMEFromTextbox(e.target.value)}),50)}function handleResize(){try{jsmeApplet?.setMenuScale(getJsmeGuiScale()),jsmeApplet?.setWidth(getJsmeContainerWidthPx())}catch(e){}}function initializeWhenReady(){"undefined"!=typeof JSApplet&&JSApplet.JSME?initializeJSME():setTimeout(initializeWhenReady,2e3)}function startInitialization(){"loading"===document.readyState?document.addEventListener("DOMContentLoaded",(()=>{setTimeout(initializeWhenReady,2e3)})):setTimeout(initializeWhenReady,2e3)}let jsmeApplet=null,lastTextboxValue="";const DEFAULT_SMILES="CCO",CONTAINER_HEIGHT="450px",SMILES_INPUT_SELECTOR="#smiles_input textarea, #smiles_input input",PASTE_DELAY=50,INIT_RETRY_DELAY=2e3,GRADIO_CHANGE_EVENTS=["input","change","keyup"];window.setJSMESmiles=function(e){return updateJSMEFromTextbox(e),updateGradioTextbox(e),e},window.clearJSME=function(){return jsmeApplet?.reset(),updateGradioTextbox(""),["","",[],[],"Cleared - Draw a new molecule or enter SMILES"]},startInitialization();