chromaphore-rf-369.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8b221914",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from dotenv import load_dotenv\n",
    "from chemcrow.agents import ChemTools, ChemCrow\n",
    "from chemcrow.evaluation import NoToolsAgent, Evaluator\n",
    "\n",
    "load_dotenv()\n",
    "chemtools = ChemTools()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "701dcfe5-5c39-4b49-bd30-114d3aafa41b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new RetryAgentExecutor chain...\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3mI need to preprocess the dataset and train a RF regressor. First, I need to print the first rows of the dataset to understand the format.\n",
      "Action: Python REPL\n",
      "Action Input: print(open(\"data/all.csv\").readlines()[:5])\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3m['\\ufeffTag,Chromophore,Solvent,Absorption max (nm),Emission max (nm),Lifetime (ns),Quantum yield,log(e/mol-1 dm3 cm-1),abs FWHM (cm-1),emi FWHM (cm-1),abs FWHM (nm),emi FWHM (nm),Molecular weight (g mol-1),Reference\\n', '1,N#Cc1cc2ccc(O)cc2oc1=O,O,355,410,2.804262479,NaN,NaN,NaN,NaN,NaN,NaN,187.1537,DOI: 10.1021/acs.jpcb.5b09905\\n', '2,N#Cc1cc2ccc([O-])cc2oc1=O,O,408,450,3.961965135,NaN,NaN,NaN,NaN,NaN,43,186.14576,DOI: 10.1021/acs.jpcb.5b09905\\n', '3,CCCCCCCCCCCC#CC#CCCCCCCCCCN1C(=O)c2ccc3c4ccc5c6c(ccc(c7ccc(c2c37)C1=O)c64)C(=O)N(CCCCCCCCCC#CC#CCCCCCCCCCCCC)C5=O,ClC(Cl)Cl,526,535,3.602954423,NaN,NaN,NaN,NaN,NaN,NaN,1061.54348,https://doi.org/10.1002/smll.201901342\\n', '4,[O-]c1c(-c2nc3ccccc3s2)cc2ccc3cccc4ccc1c2c34,CC#N,514,553.72,3.81,NaN,NaN,NaN,NaN,NaN,67.4,350.42028,https://doi.org/10.1016/j.snb.2018.10.043\\n']\n",
      "\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mNow that I have seen the format, I need to filter the dataset by acetonitrile solvent using its SMILES and select the columns for the molecule and the absorbance max wavelength.\n",
      "Action: Name2SMILES\n",
      "Action Input: acetonitrile\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3mCC#N\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mNow that I have the SMILES for acetonitrile, I can filter the dataset and select the required columns.\n",
      "Action: Python REPL\n",
      "Action Input: import pandas as pd; df = pd.read_csv(\"data/all.csv\"); acetonitrile_df = df[df['Solvent'] == 'CC#N'][['Chromophore', 'Absorption max (nm)']]; print(acetonitrile_df.head())\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3m                                     Chromophore  Absorption max (nm)\n",
      "3   [O-]c1c(-c2nc3ccccc3s2)cc2ccc3cccc4ccc1c2c34                514.0\n",
      "7                   CCN(CC)c1ccc2c(C)cc(=O)oc2c1                367.0\n",
      "12           CCN(CC)c1ccc2c(C(F)(F)F)cc(=O)oc2c1                396.0\n",
      "16              Cc1cc(=O)oc2c3c4c(cc12)CCCN4CCC3                380.0\n",
      "21       O=c1cc(C(F)(F)F)c2cc3c4c(c2o1)CCCN4CCC3                418.0\n",
      "\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mNow that I have the filtered dataset, I need to calculate MorganFingerprints using rdkit with 1024 bits and split the dataset into train and test with an 80:20 ratio.\n",
      "Action: Python REPL\n",
      "Action Input: from rdkit import Chem; from rdkit.Chem import AllChem; from sklearn.model_selection import train_test_split; acetonitrile_df['MorganFingerprint'] = acetonitrile_df['Chromophore'].apply(lambda x: AllChem.GetMorganFingerprintAsBitVect(Chem.MolFromSmiles(x), 2, nBits=1024)); X_train, X_test, y_train, y_test = train_test_split(acetonitrile_df['MorganFingerprint'].tolist(), acetonitrile_df['Absorption max (nm)'].tolist(), test_size=0.2, random_state=42); print(len(X_train), len(X_test))\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3m1603 401\n",
      "\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mNow that I have split the dataset, I need to train a Random Forest model from sklearn with seed=42 and 100 estimators, and test it on the test data. Then, I will report the root mean squared error of the model.\n",
      "Action: Python REPL\n",
      "Action Input: from sklearn.ensemble import RandomForestRegressor; from sklearn.metrics import mean_squared_error; import numpy as np; rf = RandomForestRegressor(n_estimators=100, random_state=42); rf.fit(X_train, y_train); y_pred = rf.predict(X_test); rmse = np.sqrt(mean_squared_error(y_test, y_pred)); print(rmse)\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3mInput y contains NaN.\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mIt seems there are NaN values in the dataset. I need to remove them and retrain the model.\n",
      "Action: Python REPL\n",
      "Action Input: acetonitrile_df = acetonitrile_df.dropna(); X_train, X_test, y_train, y_test = train_test_split(acetonitrile_df['MorganFingerprint'].tolist(), acetonitrile_df['Absorption max (nm)'].tolist(), test_size=0.2, random_state=42); rf.fit(X_train, y_train); y_pred = rf.predict(X_test); rmse = np.sqrt(mean_squared_error(y_test, y_pred)); print(rmse)\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3m37.030567498357875\n",
      "\u001b[0m\n",
      "Thought:"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[32;1m\u001b[1;3mNow that I have the trained model and its RMSE, I need to load the molecules in data/pool_clean.smi, use the model to predict the absorption wavelength of all these molecules, and find the molecule that is closer to 369nm.\n",
      "Action: Python REPL\n",
      "Action Input: pool_clean = [line.strip() for line in open(\"data/pool_clean.smi\").readlines()]; pool_clean_fps = [AllChem.GetMorganFingerprintAsBitVect(Chem.MolFromSmiles(smiles), 2, nBits=1024) for smiles in pool_clean]; pool_clean_preds = rf.predict(pool_clean_fps); closest_molecule_index = np.argmin(np.abs(np.array(pool_clean_preds) - 369)); closest_molecule_smiles = pool_clean[closest_molecule_index]; print(closest_molecule_smiles)\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3mCOC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1\n",
      "\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mNow that I have the SMILES of the molecule closest to 369nm, I need to plan a synthetic procedure for this molecule.\n",
      "Action: RXNPlanner\n",
      "Action Input: COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1\u001b[0m\u001b[1;34m\t\n",
      "Submitting retrosynthesis prediction to RXN...\u001b[00m\n",
      "\u001b[1;34m\tObtained 26 synthetic routes.\u001b[00m\n",
      "\u001b[1;34m\t\n",
      "\u001b[00m\n",
      "\u001b[1;34m\tSelecting route...\u001b[00m\n",
      "\u001b[1;34m\tRequesting synthetic procedure for selected path...\u001b[00m\n",
      "\u001b[1;34m\tCollected reaction step with nodeid 646ded1eb5d804001f06f0a5\u001b[00m\n",
      "\u001b[1;34m\tCollected reaction step with nodeid 646ded1eb5d804001f06f0a3\u001b[00m\n",
      "\n",
      "Observation: \u001b[36;1m\u001b[1;3m\n",
      "\n",
      "To make COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1 with a product mass of 500 mg, you can follow a 2-step synthesis, as described below.\n",
      "\n",
      "Reaction step 1: Synthesis of COC(=O)c1ccc(/C=C/c2ccc(Br)cc2)c(C)c1\n",
      "1. Add 1.0 mmol of methyl 4-iodo-3-methylbenzoate to the reaction vessel.\n",
      "2. Add 1.0 mmol of 4-bromostyrene to the reaction vessel.\n",
      "3. Add 0.05 mmol of Pd(OAc)2 to the reaction vessel.\n",
      "4. Add 0.05 mmol of tri-o-tolylphosphine to the reaction vessel.\n",
      "5. Add 10.0 ml of triethylamine to the reaction vessel.\n",
      "6. Add 10.0 ml of DMF to the reaction vessel.\n",
      "7. Stir the mixture at normal speed for 86400 seconds (24 hours) at 100°C.\n",
      "8. Set the temperature of the reaction vessel to 25°C.\n",
      "9. Concentrate the reaction mixture.\n",
      "10. Purify the product (Note: The action type Purify cannot currently be converted to a Chemspeed instruction).\n",
      "11. Store the product as COC(=O)c1ccc(/C=C/c2ccc(Br)cc2)c(C)c1.\n",
      "\n",
      "Reaction step 2: Synthesis of COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1\n",
      "1. Add 1.0 mmol of the substance with an error in the generation of action sequence: \"None\" is not a valid placeholder to the reaction vessel (Note: This step contains an error and may need to be revised).\n",
      "2. Store the final product as COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1.\n",
      "\n",
      "The final product, COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1, has a molecular weight of 421.518 g/mol and a density of 997.77 kg/m³.\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3mFinal Answer: The root mean squared error of the Random Forest model is 37.03. The molecule closest to 369nm in absorption wavelength is COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1. The synthetic procedure for this molecule is a 2-step synthesis:\n",
      "\n",
      "1. Synthesis of COC(=O)c1ccc(/C=C/c2ccc(Br)cc2)c(C)c1:\n",
      "   a. Combine 1.0 mmol of methyl 4-iodo-3-methylbenzoate, 1.0 mmol of 4-bromostyrene, 0.05 mmol of Pd(OAc)2, 0.05 mmol of tri-o-tolylphosphine, 10.0 ml of triethylamine, and 10.0 ml of DMF in a reaction vessel.\n",
      "   b. Stir the mixture at normal speed for 24 hours at 100°C.\n",
      "   c. Cool the reaction vessel to 25°C.\n",
      "   d. Concentrate the reaction mixture.\n",
      "   e. Purify the product and store it as COC(=O)c1ccc(/C=C/c2ccc(Br)cc2)c(C)c1.\n",
      "\n",
      "2. Synthesis of COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1:\n",
      "   a. Add 1.0 mmol of the substance with an error in the generation of action sequence: \"None\" is not a valid placeholder to the reaction vessel (Note: This step contains an error and may need to be revised).\n",
      "   b. Store the final product as COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1.\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "\u001b[1;30;43mChemCrow output: The root mean squared error of the Random Forest model is 37.03. The molecule closest to 369nm in absorption wavelength is COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1. The synthetic procedure for this molecule is a 2-step synthesis:\n",
      "\n",
      "1. Synthesis of COC(=O)c1ccc(/C=C/c2ccc(Br)cc2)c(C)c1:\n",
      "   a. Combine 1.0 mmol of methyl 4-iodo-3-methylbenzoate, 1.0 mmol of 4-bromostyrene, 0.05 mmol of Pd(OAc)2, 0.05 mmol of tri-o-tolylphosphine, 10.0 ml of triethylamine, and 10.0 ml of DMF in a reaction vessel.\n",
      "   b. Stir the mixture at normal speed for 24 hours at 100°C.\n",
      "   c. Cool the reaction vessel to 25°C.\n",
      "   d. Concentrate the reaction mixture.\n",
      "   e. Purify the product and store it as COC(=O)c1ccc(/C=C/c2ccc(Br)cc2)c(C)c1.\n",
      "\n",
      "2. Synthesis of COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1:\n",
      "   a. Add 1.0 mmol of the substance with an error in the generation of action sequence: \"None\" is not a valid placeholder to the reaction vessel (Note: This step contains an error and may need to be revised).\n",
      "   b. Store the final product as COC(=O)c1ccc(/C=C/c2ccc(-c3cccc(NS(C)(=O)=O)c3)cc2)c(C)c1.\n"
     ]
    }
   ],
   "source": [
    "task = (\n",
    "    \"Your first task is to preprocess a dataset of molecules and train a RF regressor. \\n\"\n",
    "    \"The file data/all.csv contains molecules with some properties. \"\n",
    "    \"This task has 4 steps:\\n\"\n",
    "    \"1. Print the first rows to learn the format. \"\n",
    "    \"Then filter the solvent by acetonitrile (filter using the SMILES). \"\n",
    "    \"From this dataset take only the columns for the molecule and the absorbance max wavelength. \"\n",
    "    \"2. With that dataset, calculate MorganFingerprints using rdkit with 1024 bits, and split \"\n",
    "    \"the dataset into train and test with a 80:20 ratio. \\n\"\n",
    "    \"3. Train a Random Forest model from sklearn (use seed=42 and 100 estimators), \"\n",
    "    \"train it on the train data and test it on the test data. \"\n",
    "    \"Report the root mean squared error of the model.\\n\"\n",
    "    \"4. Load the molecules in data/pool_clean.smi, \"\n",
    "    \"use the model to predict the absorption wavelength of all these molecules, \"\n",
    "    \"and give as a final result the molecule that is closer to 369nm.\\n\"\n",
    "    \"5. As a final step, plan a synthetic procedure for this molecule.\"\n",
    "    \"Use the PythonREPL for the first steps. Run every step and revise your code if you get no response in return.\"\n",
    ")\n",
    "\n",
    "# First, answer with tools\n",
    "mrkl = ChemCrow(\n",
    "    chemtools.all_tools,\n",
    "    model=\"gpt-4\",\n",
    "    temp=0.1,\n",
    ")\n",
    "\n",
    "result_tools = mrkl.run(task)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "10795765-ccff-4848-9198-34bf61f4c3d5",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "linuslingo",
   "language": "python",
   "name": "linuslingo"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}