空间站广场

论文

Notebooks

比赛

课程

Apps

我的主页

我的Notebooks

我的论文库

我的足迹

我的工作空间

任务

节点

文件

数据集

镜像

项目

数据库

公开

Uni-Mol Predicts Solubility of Flow Battery

Uni-Mol

Yani Guan

更新于 2024-10-18

推荐镜像 :Third-party software:Deprecated

推荐机型 :c3_m4_1 * NVIDIA T4

数据集

液流电池溶解度(v2)

Uni-Mol Predicts Solubility of Flow Battery

©️ Copyright 2023 @ Authors
Authors: Qiming Xie 📨 Mingzheng Zhang 📨
Date: 2023-09-06
Sharing Agreement: This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Quick Start: Click the Start Connection button above, select the unimol-qsar:v0.1 image and any GPU node configuration, and wait a moment to run.

代码

文本

AIMS:

Practical use of Uni-Mol for predicting solubility in redox flow batteries
Understand the working modules of Uni-Mol
Use a training method with SMILES strings as input
Use regression models to predict continuous values
Predict the solubility of some aqueous redox-active molecules in redox flow batteries using the trained model

代码

文本

Overview

1. Introduction

Aqueous organic redox flow batteries (AORFBs) show great potential as a large-scale energy storage technology, enabling efficient use of renewable energy. AORFBs use water-soluble organic redox-active molecules as electrolytes. These organic electrolytes have advantages such as abundant raw material sources, highly tunable performance, and fast kinetics, offering broader application prospects for redox flow batteries. However, organic redox-active molecules have limited solubility, electrolyte freezing, and slow reaction kinetics, which makes it challenging to maintain the performance of AORFBs under low temperatures [1].

2. Core Indicators of Redox Flow Batteries

The most critical performance indicator of redox flow batteries is energy density. The theoretical energy density of RFBs is derived from the following equation [2]:

$W = \frac{n CF V}{μ _{v}}$

Where: W is the energy density, Wh·L^-1; n is the number of electrons participating in the redox reaction; C is the minimum concentration of redox-active species dissolved in the electrolyte, mol·L^-1; F is the Faraday constant; V is the battery voltage, V; $μ_{v}$ is the volume factor, $μ_{v}$ =1+(concentration of the less soluble active species/concentration of the more soluble active species), and when concentrations are equal, $μ_{v}$ =2.

Therefore, the energy density is determined by the solubility of the active species in the electrolyte, the redox potential, and the number of transferred electrons. Consequently, the main approaches to increasing energy density focus on increasing solubility, expanding the electrochemical window, and increasing the number of electron transfers. Combining these approaches with high-throughput computational methods to screen organic redox-active molecules can facilitate the successful development of the next generation of redox flow batteries.

3. Introduction to the RedDB Database

This project uses the RedDB [3] database, which contains 31,618 molecular data entries derived from the structural functionalization of 52 different core molecules. It includes results from two types of DFT calculations: OPT geometry optimization and SPE single-point energy, as well as solubility data predicted by the water solubility prediction model (AqSolPred v1.0) [4]. The solubility values in RedDB are expressed as dimensionless logarithms in mol/L, denoted as LogS.

4. Introduction to AqSolPred

AqSolPred integrates three machine learning methods and is trained on a database of 9,982 experimental solubility data points—AqSolDB. The model has a mean absolute error (MAE) of 0.348. AqSolPred has been previously validated on benchmark solubility datasets [6]. Its MAE of 0.348 LogS is lower than that of traditional cheminformatics and machine learning methods typically used for predicting chemical solubility in water [4]. The SMILES representation of molecules is used as input for AqSolPred, and the ML-predicted solubility data has been incorporated into RedDB.

5. Objectives

Use a training method with SMILES strings as input;
Use the Uni-Mol universal molecular learning framework to predict continuous values;
Use the trained model to predict the solubility of some aqueous redox-active compounds in redox flow batteries.

代码

文本

Step1:import Uni-Mol

import Uni-Mol's modules

代码

文本

[1]

# import Uni-Mol

from unimol import MolTrain, MolPredict

代码

文本

Step 2: Input Data and Train (Optional)

Training may take a long time. The trained data (CSV files) have already been uploaded to the dataset, so those with limited time can skip this step.

代码

文本

[2]

clf = MolTrain(task='regression',

data_type='molecule',

epochs=50,

learning_rate=0.0004,

batch_size=32,

early_stopping=5,

metrics='r2',

split='random',

save_path='./best_exp',

)

clf.fit('/bohr/aorfb-7iz1/v2/RedDB_train.csv')

2023-09-04 15:51:18 | unimol/data/datareader.py | 138 | INFO | Uni-Mol(QSAR) | Anomaly clean with 3 sigma threshold: 25294 -> 25293
2023-09-04 15:51:22 | unimol/data/conformer.py | 62 | INFO | Uni-Mol(QSAR) | Start generating conformers...
25293it [02:46, 151.98it/s]
2023-09-04 15:54:09 | unimol/data/conformer.py | 66 | INFO | Uni-Mol(QSAR) | Failed to generate conformers for 0.00% of molecules.
2023-09-04 15:54:09 | unimol/data/conformer.py | 68 | INFO | Uni-Mol(QSAR) | Failed to generate 3d conformers for 0.03% of molecules.
2023-09-04 15:54:09 | unimol/train.py | 88 | INFO | Uni-Mol(QSAR) | Output directory already exists: ./best_exp
2023-09-04 15:54:09 | unimol/train.py | 89 | INFO | Uni-Mol(QSAR) | Warning: Overwrite output directory: ./best_exp
2023-09-04 15:54:10 | unimol/models/unimol.py | 114 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /opt/conda/lib/python3.8/site-packages/unimol-0.0.2-py3.8.egg/unimol/weights/mol_pre_all_h_220816.pt
2023-09-04 15:54:10 | unimol/models/nnmodel.py | 103 | INFO | Uni-Mol(QSAR) | start training Uni-Mol:unimolv1
2023-09-04 15:55:22 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [1/50] train_loss: 0.1694, val_loss: 0.0961, val_r2: 0.9043, lr: 0.000267, 69.2s
2023-09-04 15:56:27 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [2/50] train_loss: 0.0543, val_loss: 0.0285, val_r2: 0.9716, lr: 0.000396, 63.7s
2023-09-04 15:57:32 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [3/50] train_loss: 0.0273, val_loss: 0.0284, val_r2: 0.9716, lr: 0.000388, 64.2s
2023-09-04 15:58:36 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [4/50] train_loss: 0.0210, val_loss: 0.0169, val_r2: 0.9831, lr: 0.000379, 63.1s
2023-09-04 15:59:44 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [5/50] train_loss: 0.0184, val_loss: 0.0140, val_r2: 0.9860, lr: 0.000371, 67.6s
2023-09-04 16:00:56 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [6/50] train_loss: 0.0151, val_loss: 0.0165, val_r2: 0.9835, lr: 0.000363, 71.0s
2023-09-04 16:01:59 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [7/50] train_loss: 0.0144, val_loss: 0.0121, val_r2: 0.9879, lr: 0.000355, 63.4s
2023-09-04 16:03:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [8/50] train_loss: 0.0126, val_loss: 0.0121, val_r2: 0.9878, lr: 0.000346, 63.7s
2023-09-04 16:04:07 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [9/50] train_loss: 0.0115, val_loss: 0.0169, val_r2: 0.9831, lr: 0.000338, 63.7s
2023-09-04 16:05:13 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [10/50] train_loss: 0.0103, val_loss: 0.0141, val_r2: 0.9858, lr: 0.000330, 65.6s
2023-09-04 16:06:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [11/50] train_loss: 0.0102, val_loss: 0.0122, val_r2: 0.9877, lr: 0.000322, 66.1s
2023-09-04 16:07:21 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [12/50] train_loss: 0.0092, val_loss: 0.0084, val_r2: 0.9917, lr: 0.000313, 62.2s
2023-09-04 16:08:26 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [13/50] train_loss: 0.0084, val_loss: 0.0087, val_r2: 0.9913, lr: 0.000305, 64.6s
2023-09-04 16:09:37 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [14/50] train_loss: 0.0074, val_loss: 0.0092, val_r2: 0.9908, lr: 0.000297, 70.3s
2023-09-04 16:10:44 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [15/50] train_loss: 0.0081, val_loss: 0.0113, val_r2: 0.9887, lr: 0.000289, 67.2s
2023-09-04 16:11:53 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [16/50] train_loss: 0.0069, val_loss: 0.0080, val_r2: 0.9920, lr: 0.000280, 69.3s
2023-09-04 16:12:57 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [17/50] train_loss: 0.0064, val_loss: 0.0090, val_r2: 0.9909, lr: 0.000272, 62.8s
2023-09-04 16:14:08 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [18/50] train_loss: 0.0057, val_loss: 0.0069, val_r2: 0.9931, lr: 0.000264, 71.6s
2023-09-04 16:15:12 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [19/50] train_loss: 0.0054, val_loss: 0.0069, val_r2: 0.9931, lr: 0.000256, 63.6s
2023-09-04 16:16:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [20/50] train_loss: 0.0056, val_loss: 0.0074, val_r2: 0.9926, lr: 0.000247, 65.7s
2023-09-04 16:17:21 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [21/50] train_loss: 0.0050, val_loss: 0.0071, val_r2: 0.9929, lr: 0.000239, 61.8s
2023-09-04 16:18:24 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [22/50] train_loss: 0.0046, val_loss: 0.0085, val_r2: 0.9916, lr: 0.000231, 63.6s
2023-09-04 16:19:29 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [23/50] train_loss: 0.0041, val_loss: 0.0061, val_r2: 0.9939, lr: 0.000223, 64.9s
2023-09-04 16:20:37 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [24/50] train_loss: 0.0040, val_loss: 0.0062, val_r2: 0.9938, lr: 0.000214, 67.2s
2023-09-04 16:21:41 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [25/50] train_loss: 0.0043, val_loss: 0.0061, val_r2: 0.9939, lr: 0.000206, 64.2s
2023-09-04 16:22:45 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [26/50] train_loss: 0.0037, val_loss: 0.0056, val_r2: 0.9944, lr: 0.000198, 63.6s
2023-09-04 16:23:57 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [27/50] train_loss: 0.0035, val_loss: 0.0057, val_r2: 0.9943, lr: 0.000190, 71.6s
2023-09-04 16:25:01 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [28/50] train_loss: 0.0033, val_loss: 0.0058, val_r2: 0.9942, lr: 0.000181, 64.2s
2023-09-04 16:26:06 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [29/50] train_loss: 0.0031, val_loss: 0.0056, val_r2: 0.9944, lr: 0.000173, 65.1s
2023-09-04 16:27:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [30/50] train_loss: 0.0030, val_loss: 0.0055, val_r2: 0.9945, lr: 0.000165, 72.3s
2023-09-04 16:28:31 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [31/50] train_loss: 0.0029, val_loss: 0.0049, val_r2: 0.9950, lr: 0.000157, 72.2s
2023-09-04 16:29:40 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [32/50] train_loss: 0.0028, val_loss: 0.0051, val_r2: 0.9949, lr: 0.000148, 67.9s
2023-09-04 16:30:45 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [33/50] train_loss: 0.0026, val_loss: 0.0051, val_r2: 0.9949, lr: 0.000140, 65.1s
2023-09-04 16:31:48 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [34/50] train_loss: 0.0026, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000132, 62.8s
2023-09-04 16:33:04 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [35/50] train_loss: 0.0023, val_loss: 0.0049, val_r2: 0.9951, lr: 0.000124, 76.5s
2023-09-04 16:34:22 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [36/50] train_loss: 0.0022, val_loss: 0.0048, val_r2: 0.9952, lr: 0.000115, 76.7s
2023-09-04 16:35:30 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [37/50] train_loss: 0.0021, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000107, 67.9s
2023-09-04 16:36:41 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [38/50] train_loss: 0.0020, val_loss: 0.0048, val_r2: 0.9951, lr: 0.000099, 70.6s
2023-09-04 16:37:53 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [39/50] train_loss: 0.0019, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000091, 72.5s
2023-09-04 16:38:59 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [40/50] train_loss: 0.0020, val_loss: 0.0045, val_r2: 0.9955, lr: 0.000082, 64.9s
2023-09-04 16:40:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [41/50] train_loss: 0.0018, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000074, 63.6s
2023-09-04 16:41:07 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [42/50] train_loss: 0.0018, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000066, 64.4s
2023-09-04 16:42:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [43/50] train_loss: 0.0018, val_loss: 0.0045, val_r2: 0.9955, lr: 0.000058, 62.9s
2023-09-04 16:43:14 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [44/50] train_loss: 0.0016, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000049, 63.2s
2023-09-04 16:44:16 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [45/50] train_loss: 0.0015, val_loss: 0.0045, val_r2: 0.9954, lr: 0.000041, 62.2s
2023-09-04 16:45:20 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [46/50] train_loss: 0.0015, val_loss: 0.0044, val_r2: 0.9956, lr: 0.000033, 63.6s
2023-09-04 16:46:29 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [47/50] train_loss: 0.0014, val_loss: 0.0044, val_r2: 0.9956, lr: 0.000025, 68.4s
2023-09-04 16:47:33 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [48/50] train_loss: 0.0014, val_loss: 0.0044, val_r2: 0.9956, lr: 0.000016, 64.3s
2023-09-04 16:48:37 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [49/50] train_loss: 0.0014, val_loss: 0.0045, val_r2: 0.9955, lr: 0.000008, 63.3s
2023-09-04 16:49:40 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [50/50] train_loss: 0.0013, val_loss: 0.0044, val_r2: 0.9956, lr: 0.000000, 63.5s
2023-09-04 16:49:41 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-04 16:49:46 | unimol/models/nnmodel.py | 129 | INFO | Uni-Mol(QSAR) | fold 0, result {'r2': 0.9956226848041148, 'mae': 0.05991403, 'pearsonr': 0.9978418748820779, 'spearmanr': 0.9973003817239617, 'mse': 0.0075112535}
2023-09-04 16:49:46 | unimol/models/unimol.py | 114 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /opt/conda/lib/python3.8/site-packages/unimol-0.0.2-py3.8.egg/unimol/weights/mol_pre_all_h_220816.pt
2023-09-04 16:50:51 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [1/50] train_loss: 0.1861, val_loss: 0.0841, val_r2: 0.9168, lr: 0.000267, 64.9s
2023-09-04 16:51:57 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [2/50] train_loss: 0.0599, val_loss: 0.0362, val_r2: 0.9642, lr: 0.000396, 64.7s
2023-09-04 16:53:06 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [3/50] train_loss: 0.0300, val_loss: 0.0262, val_r2: 0.9741, lr: 0.000388, 68.0s
2023-09-04 16:54:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [4/50] train_loss: 0.0227, val_loss: 0.0228, val_r2: 0.9774, lr: 0.000379, 63.1s
2023-09-04 16:55:22 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [5/50] train_loss: 0.0179, val_loss: 0.0199, val_r2: 0.9804, lr: 0.000371, 71.4s
2023-09-04 16:56:29 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [6/50] train_loss: 0.0166, val_loss: 0.0142, val_r2: 0.9859, lr: 0.000363, 66.7s
2023-09-04 16:57:33 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [7/50] train_loss: 0.0142, val_loss: 0.0163, val_r2: 0.9839, lr: 0.000355, 63.5s
2023-09-04 16:58:37 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [8/50] train_loss: 0.0136, val_loss: 0.0163, val_r2: 0.9839, lr: 0.000346, 64.0s
2023-09-04 16:59:43 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [9/50] train_loss: 0.0113, val_loss: 0.0192, val_r2: 0.9810, lr: 0.000338, 66.1s
2023-09-04 17:00:46 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [10/50] train_loss: 0.0119, val_loss: 0.0127, val_r2: 0.9874, lr: 0.000330, 62.6s
2023-09-04 17:01:50 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [11/50] train_loss: 0.0110, val_loss: 0.0165, val_r2: 0.9836, lr: 0.000322, 63.0s
2023-09-04 17:02:54 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [12/50] train_loss: 0.0105, val_loss: 0.0100, val_r2: 0.9901, lr: 0.000313, 64.6s
2023-09-04 17:04:02 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [13/50] train_loss: 0.0091, val_loss: 0.0092, val_r2: 0.9909, lr: 0.000305, 67.8s
2023-09-04 17:05:06 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [14/50] train_loss: 0.0097, val_loss: 0.0091, val_r2: 0.9910, lr: 0.000297, 63.1s
2023-09-04 17:06:11 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [15/50] train_loss: 0.0076, val_loss: 0.0077, val_r2: 0.9923, lr: 0.000289, 63.6s
2023-09-04 17:07:14 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [16/50] train_loss: 0.0072, val_loss: 0.0110, val_r2: 0.9891, lr: 0.000280, 62.7s
2023-09-04 17:08:17 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [17/50] train_loss: 0.0075, val_loss: 0.0071, val_r2: 0.9929, lr: 0.000272, 62.6s
2023-09-04 17:09:21 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [18/50] train_loss: 0.0065, val_loss: 0.0072, val_r2: 0.9929, lr: 0.000264, 63.8s
2023-09-04 17:10:27 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [19/50] train_loss: 0.0063, val_loss: 0.0086, val_r2: 0.9914, lr: 0.000256, 65.5s
2023-09-04 17:11:30 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [20/50] train_loss: 0.0064, val_loss: 0.0078, val_r2: 0.9922, lr: 0.000247, 63.8s
2023-09-04 17:12:33 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [21/50] train_loss: 0.0055, val_loss: 0.0058, val_r2: 0.9943, lr: 0.000239, 63.1s
2023-09-04 17:13:40 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [22/50] train_loss: 0.0052, val_loss: 0.0063, val_r2: 0.9938, lr: 0.000231, 66.2s
2023-09-04 17:14:49 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [23/50] train_loss: 0.0053, val_loss: 0.0090, val_r2: 0.9911, lr: 0.000223, 69.1s
2023-09-04 17:16:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [24/50] train_loss: 0.0051, val_loss: 0.0058, val_r2: 0.9943, lr: 0.000214, 73.9s
2023-09-04 17:17:08 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [25/50] train_loss: 0.0047, val_loss: 0.0053, val_r2: 0.9947, lr: 0.000206, 63.8s
2023-09-04 17:18:12 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [26/50] train_loss: 0.0040, val_loss: 0.0056, val_r2: 0.9945, lr: 0.000198, 63.4s
2023-09-04 17:19:18 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [27/50] train_loss: 0.0040, val_loss: 0.0060, val_r2: 0.9941, lr: 0.000190, 65.5s
2023-09-04 17:20:20 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [28/50] train_loss: 0.0039, val_loss: 0.0070, val_r2: 0.9931, lr: 0.000181, 62.9s
2023-09-04 17:21:35 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [29/50] train_loss: 0.0036, val_loss: 0.0080, val_r2: 0.9921, lr: 0.000173, 74.9s
2023-09-04 17:22:39 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [30/50] train_loss: 0.0036, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000165, 64.1s
2023-09-04 17:23:44 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [31/50] train_loss: 0.0036, val_loss: 0.0060, val_r2: 0.9941, lr: 0.000157, 63.5s
2023-09-04 17:24:48 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [32/50] train_loss: 0.0033, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000148, 64.0s
2023-09-04 17:25:51 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [33/50] train_loss: 0.0032, val_loss: 0.0044, val_r2: 0.9956, lr: 0.000140, 63.4s
2023-09-04 17:27:01 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [34/50] train_loss: 0.0028, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000132, 69.0s
2023-09-04 17:28:11 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [35/50] train_loss: 0.0026, val_loss: 0.0043, val_r2: 0.9958, lr: 0.000124, 70.3s
2023-09-04 17:29:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [36/50] train_loss: 0.0025, val_loss: 0.0052, val_r2: 0.9949, lr: 0.000115, 67.5s
2023-09-04 17:30:25 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [37/50] train_loss: 0.0025, val_loss: 0.0046, val_r2: 0.9955, lr: 0.000107, 65.6s
2023-09-04 17:31:30 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [38/50] train_loss: 0.0024, val_loss: 0.0046, val_r2: 0.9955, lr: 0.000099, 65.2s
2023-09-04 17:32:34 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [39/50] train_loss: 0.0023, val_loss: 0.0042, val_r2: 0.9958, lr: 0.000091, 63.9s
2023-09-04 17:33:39 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [40/50] train_loss: 0.0022, val_loss: 0.0054, val_r2: 0.9947, lr: 0.000082, 64.3s
2023-09-04 17:34:44 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [41/50] train_loss: 0.0020, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000074, 64.5s
2023-09-04 17:35:48 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [42/50] train_loss: 0.0020, val_loss: 0.0049, val_r2: 0.9951, lr: 0.000066, 64.6s
2023-09-04 17:36:56 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [43/50] train_loss: 0.0018, val_loss: 0.0047, val_r2: 0.9953, lr: 0.000058, 67.7s
2023-09-04 17:38:05 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [44/50] train_loss: 0.0018, val_loss: 0.0041, val_r2: 0.9959, lr: 0.000049, 68.6s
2023-09-04 17:39:14 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [45/50] train_loss: 0.0017, val_loss: 0.0042, val_r2: 0.9959, lr: 0.000041, 68.0s
2023-09-04 17:40:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [46/50] train_loss: 0.0017, val_loss: 0.0040, val_r2: 0.9960, lr: 0.000033, 65.3s
2023-09-04 17:41:23 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [47/50] train_loss: 0.0019, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000025, 63.4s
2023-09-04 17:42:32 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [48/50] train_loss: 0.0019, val_loss: 0.0042, val_r2: 0.9959, lr: 0.000016, 68.8s
2023-09-04 17:43:37 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [49/50] train_loss: 0.0015, val_loss: 0.0040, val_r2: 0.9960, lr: 0.000008, 64.9s
2023-09-04 17:44:43 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [50/50] train_loss: 0.0015, val_loss: 0.0042, val_r2: 0.9959, lr: 0.000000, 66.0s
2023-09-04 17:44:45 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-04 17:44:50 | unimol/models/nnmodel.py | 129 | INFO | Uni-Mol(QSAR) | fold 1, result {'r2': 0.9960434374756032, 'mae': 0.059422977, 'pearsonr': 0.9980879792930252, 'spearmanr': 0.997501112441128, 'mse': 0.0068540904}
2023-09-04 17:44:50 | unimol/models/unimol.py | 114 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /opt/conda/lib/python3.8/site-packages/unimol-0.0.2-py3.8.egg/unimol/weights/mol_pre_all_h_220816.pt
2023-09-04 17:45:54 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [1/50] train_loss: 0.1815, val_loss: 0.0471, val_r2: 0.9540, lr: 0.000267, 64.0s
2023-09-04 17:47:01 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [2/50] train_loss: 0.0518, val_loss: 0.0225, val_r2: 0.9778, lr: 0.000396, 65.7s
2023-09-04 17:48:20 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [3/50] train_loss: 0.0276, val_loss: 0.0249, val_r2: 0.9756, lr: 0.000388, 77.8s
2023-09-04 17:49:31 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [4/50] train_loss: 0.0196, val_loss: 0.0150, val_r2: 0.9852, lr: 0.000379, 71.6s
2023-09-04 17:50:45 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [5/50] train_loss: 0.0169, val_loss: 0.0158, val_r2: 0.9843, lr: 0.000371, 73.1s
2023-09-04 17:51:53 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [6/50] train_loss: 0.0153, val_loss: 0.0114, val_r2: 0.9888, lr: 0.000363, 68.1s
2023-09-04 17:52:59 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [7/50] train_loss: 0.0131, val_loss: 0.0123, val_r2: 0.9880, lr: 0.000355, 65.7s
2023-09-04 17:54:04 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [8/50] train_loss: 0.0123, val_loss: 0.0099, val_r2: 0.9902, lr: 0.000346, 64.8s
2023-09-04 17:55:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [9/50] train_loss: 0.0105, val_loss: 0.0110, val_r2: 0.9891, lr: 0.000338, 64.9s
2023-09-04 17:56:12 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [10/50] train_loss: 0.0101, val_loss: 0.0123, val_r2: 0.9878, lr: 0.000330, 62.7s
2023-09-04 17:57:16 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [11/50] train_loss: 0.0094, val_loss: 0.0116, val_r2: 0.9885, lr: 0.000322, 63.1s
2023-09-04 17:58:18 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [12/50] train_loss: 0.0089, val_loss: 0.0103, val_r2: 0.9899, lr: 0.000313, 62.9s
2023-09-04 17:59:22 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [13/50] train_loss: 0.0080, val_loss: 0.0082, val_r2: 0.9919, lr: 0.000305, 63.6s
2023-09-04 18:00:28 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [14/50] train_loss: 0.0081, val_loss: 0.0089, val_r2: 0.9912, lr: 0.000297, 65.5s
2023-09-04 18:01:31 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [15/50] train_loss: 0.0072, val_loss: 0.0107, val_r2: 0.9894, lr: 0.000289, 63.3s
2023-09-04 18:02:34 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [16/50] train_loss: 0.0068, val_loss: 0.0074, val_r2: 0.9927, lr: 0.000280, 62.6s
2023-09-04 18:03:38 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [17/50] train_loss: 0.0060, val_loss: 0.0069, val_r2: 0.9932, lr: 0.000272, 63.2s
2023-09-04 18:04:45 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [18/50] train_loss: 0.0055, val_loss: 0.0074, val_r2: 0.9927, lr: 0.000264, 66.7s
2023-09-04 18:05:56 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [19/50] train_loss: 0.0054, val_loss: 0.0067, val_r2: 0.9933, lr: 0.000256, 70.5s
2023-09-04 18:06:59 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [20/50] train_loss: 0.0053, val_loss: 0.0072, val_r2: 0.9929, lr: 0.000247, 62.5s
2023-09-04 18:08:01 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [21/50] train_loss: 0.0049, val_loss: 0.0074, val_r2: 0.9927, lr: 0.000239, 62.2s
2023-09-04 18:09:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [22/50] train_loss: 0.0044, val_loss: 0.0063, val_r2: 0.9938, lr: 0.000231, 62.6s
2023-09-04 18:10:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [23/50] train_loss: 0.0045, val_loss: 0.0072, val_r2: 0.9929, lr: 0.000223, 66.2s
2023-09-04 18:11:14 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [24/50] train_loss: 0.0041, val_loss: 0.0057, val_r2: 0.9943, lr: 0.000214, 64.2s
2023-09-04 18:12:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [25/50] train_loss: 0.0043, val_loss: 0.0069, val_r2: 0.9932, lr: 0.000206, 63.9s
2023-09-04 18:13:22 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [26/50] train_loss: 0.0038, val_loss: 0.0057, val_r2: 0.9944, lr: 0.000198, 63.1s
2023-09-04 18:14:27 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [27/50] train_loss: 0.0034, val_loss: 0.0054, val_r2: 0.9946, lr: 0.000190, 64.9s
2023-09-04 18:15:46 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [28/50] train_loss: 0.0034, val_loss: 0.0060, val_r2: 0.9941, lr: 0.000181, 78.0s
2023-09-04 18:17:04 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [29/50] train_loss: 0.0032, val_loss: 0.0054, val_r2: 0.9947, lr: 0.000173, 77.9s
2023-09-04 18:18:17 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [30/50] train_loss: 0.0029, val_loss: 0.0053, val_r2: 0.9947, lr: 0.000165, 72.7s
2023-09-04 18:19:21 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [31/50] train_loss: 0.0030, val_loss: 0.0053, val_r2: 0.9948, lr: 0.000157, 63.4s
2023-09-04 18:20:27 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [32/50] train_loss: 0.0026, val_loss: 0.0050, val_r2: 0.9951, lr: 0.000148, 64.5s
2023-09-04 18:21:32 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [33/50] train_loss: 0.0024, val_loss: 0.0053, val_r2: 0.9948, lr: 0.000140, 64.2s
2023-09-04 18:22:35 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [34/50] train_loss: 0.0025, val_loss: 0.0047, val_r2: 0.9954, lr: 0.000132, 63.0s
2023-09-04 18:23:42 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [35/50] train_loss: 0.0023, val_loss: 0.0047, val_r2: 0.9954, lr: 0.000124, 67.0s
2023-09-04 18:24:47 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [36/50] train_loss: 0.0022, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000115, 64.3s
2023-09-04 18:25:58 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [37/50] train_loss: 0.0020, val_loss: 0.0045, val_r2: 0.9956, lr: 0.000107, 70.8s
2023-09-04 18:27:01 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [38/50] train_loss: 0.0020, val_loss: 0.0046, val_r2: 0.9955, lr: 0.000099, 62.5s
2023-09-04 18:28:05 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [39/50] train_loss: 0.0018, val_loss: 0.0046, val_r2: 0.9955, lr: 0.000091, 63.7s
2023-09-04 18:29:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [40/50] train_loss: 0.0018, val_loss: 0.0047, val_r2: 0.9953, lr: 0.000082, 64.7s
2023-09-04 18:30:12 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [41/50] train_loss: 0.0018, val_loss: 0.0043, val_r2: 0.9958, lr: 0.000074, 62.3s
2023-09-04 18:31:15 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [42/50] train_loss: 0.0017, val_loss: 0.0045, val_r2: 0.9956, lr: 0.000066, 62.5s
2023-09-04 18:32:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [43/50] train_loss: 0.0016, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000058, 64.3s
2023-09-04 18:33:24 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [44/50] train_loss: 0.0016, val_loss: 0.0043, val_r2: 0.9958, lr: 0.000049, 65.1s
2023-09-04 18:34:29 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [45/50] train_loss: 0.0017, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000041, 64.3s
2023-09-04 18:35:36 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [46/50] train_loss: 0.0014, val_loss: 0.0042, val_r2: 0.9958, lr: 0.000033, 67.2s
2023-09-04 18:36:47 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [47/50] train_loss: 0.0013, val_loss: 0.0044, val_r2: 0.9957, lr: 0.000025, 69.9s
2023-09-04 18:38:05 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [48/50] train_loss: 0.0013, val_loss: 0.0042, val_r2: 0.9958, lr: 0.000016, 78.2s
2023-09-04 18:39:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [49/50] train_loss: 0.0014, val_loss: 0.0042, val_r2: 0.9959, lr: 0.000008, 73.7s
2023-09-04 18:40:22 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [50/50] train_loss: 0.0013, val_loss: 0.0042, val_r2: 0.9959, lr: 0.000000, 62.9s
2023-09-04 18:40:23 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-04 18:40:28 | unimol/models/nnmodel.py | 129 | INFO | Uni-Mol(QSAR) | fold 2, result {'r2': 0.9959009040119442, 'mae': 0.058838677, 'pearsonr': 0.9979535268752111, 'spearmanr': 0.9973975017714561, 'mse': 0.007119618}
2023-09-04 18:40:29 | unimol/models/unimol.py | 114 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /opt/conda/lib/python3.8/site-packages/unimol-0.0.2-py3.8.egg/unimol/weights/mol_pre_all_h_220816.pt
2023-09-04 18:41:33 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [1/50] train_loss: 0.1751, val_loss: 0.0821, val_r2: 0.9176, lr: 0.000267, 64.2s
2023-09-04 18:42:36 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [2/50] train_loss: 0.0520, val_loss: 0.0254, val_r2: 0.9746, lr: 0.000396, 62.4s
2023-09-04 18:43:42 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [3/50] train_loss: 0.0273, val_loss: 0.0234, val_r2: 0.9765, lr: 0.000388, 64.9s
2023-09-04 18:44:53 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [4/50] train_loss: 0.0211, val_loss: 0.0153, val_r2: 0.9847, lr: 0.000379, 71.2s
2023-09-04 18:46:04 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [5/50] train_loss: 0.0181, val_loss: 0.0128, val_r2: 0.9873, lr: 0.000371, 70.1s
2023-09-04 18:47:11 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [6/50] train_loss: 0.0152, val_loss: 0.0138, val_r2: 0.9861, lr: 0.000363, 66.3s
2023-09-04 18:48:15 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [7/50] train_loss: 0.0132, val_loss: 0.0132, val_r2: 0.9868, lr: 0.000355, 64.3s
2023-09-04 18:49:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [8/50] train_loss: 0.0130, val_loss: 0.0111, val_r2: 0.9889, lr: 0.000346, 64.0s
2023-09-04 18:50:27 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [9/50] train_loss: 0.0123, val_loss: 0.0120, val_r2: 0.9881, lr: 0.000338, 66.6s
2023-09-04 18:51:32 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [10/50] train_loss: 0.0108, val_loss: 0.0095, val_r2: 0.9904, lr: 0.000330, 65.3s
2023-09-04 18:52:35 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [11/50] train_loss: 0.0097, val_loss: 0.0084, val_r2: 0.9916, lr: 0.000322, 62.6s
2023-09-04 18:53:38 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [12/50] train_loss: 0.0087, val_loss: 0.0084, val_r2: 0.9916, lr: 0.000313, 62.6s
2023-09-04 18:54:42 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [13/50] train_loss: 0.0086, val_loss: 0.0099, val_r2: 0.9901, lr: 0.000305, 62.7s
2023-09-04 18:55:47 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [14/50] train_loss: 0.0076, val_loss: 0.0084, val_r2: 0.9916, lr: 0.000297, 65.1s
2023-09-04 18:56:52 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [15/50] train_loss: 0.0076, val_loss: 0.0079, val_r2: 0.9921, lr: 0.000289, 64.3s
2023-09-04 18:58:07 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [16/50] train_loss: 0.0068, val_loss: 0.0074, val_r2: 0.9926, lr: 0.000280, 74.5s
2023-09-04 18:59:26 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [17/50] train_loss: 0.0062, val_loss: 0.0077, val_r2: 0.9923, lr: 0.000272, 78.4s
2023-09-04 19:00:44 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [18/50] train_loss: 0.0060, val_loss: 0.0068, val_r2: 0.9931, lr: 0.000264, 78.2s
2023-09-04 19:02:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [19/50] train_loss: 0.0060, val_loss: 0.0065, val_r2: 0.9935, lr: 0.000256, 78.1s
2023-09-04 19:03:12 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [20/50] train_loss: 0.0056, val_loss: 0.0081, val_r2: 0.9919, lr: 0.000247, 67.9s
2023-09-04 19:04:17 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [21/50] train_loss: 0.0047, val_loss: 0.0073, val_r2: 0.9926, lr: 0.000239, 65.1s
2023-09-04 19:05:20 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [22/50] train_loss: 0.0053, val_loss: 0.0075, val_r2: 0.9924, lr: 0.000231, 63.3s
2023-09-04 19:06:23 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [23/50] train_loss: 0.0046, val_loss: 0.0060, val_r2: 0.9940, lr: 0.000223, 63.3s
2023-09-04 19:07:29 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [24/50] train_loss: 0.0040, val_loss: 0.0057, val_r2: 0.9943, lr: 0.000214, 65.0s
2023-09-04 19:08:36 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [25/50] train_loss: 0.0038, val_loss: 0.0068, val_r2: 0.9931, lr: 0.000206, 65.9s
2023-09-04 19:09:42 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [26/50] train_loss: 0.0039, val_loss: 0.0056, val_r2: 0.9944, lr: 0.000198, 66.0s
2023-09-04 19:10:49 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [27/50] train_loss: 0.0034, val_loss: 0.0055, val_r2: 0.9944, lr: 0.000190, 65.6s
2023-09-04 19:11:56 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [28/50] train_loss: 0.0035, val_loss: 0.0064, val_r2: 0.9936, lr: 0.000181, 66.5s
2023-09-04 19:12:58 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [29/50] train_loss: 0.0033, val_loss: 0.0067, val_r2: 0.9933, lr: 0.000173, 62.4s
2023-09-04 19:14:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [30/50] train_loss: 0.0031, val_loss: 0.0053, val_r2: 0.9947, lr: 0.000165, 64.5s
2023-09-04 19:15:07 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [31/50] train_loss: 0.0029, val_loss: 0.0048, val_r2: 0.9952, lr: 0.000157, 63.0s
2023-09-04 19:16:11 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [32/50] train_loss: 0.0027, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000148, 63.7s
2023-09-04 19:17:29 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [33/50] train_loss: 0.0025, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000140, 77.8s
2023-09-04 19:18:42 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [34/50] train_loss: 0.0026, val_loss: 0.0055, val_r2: 0.9945, lr: 0.000132, 72.9s
2023-09-04 19:19:55 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [35/50] train_loss: 0.0024, val_loss: 0.0048, val_r2: 0.9952, lr: 0.000124, 73.3s
2023-09-04 19:21:05 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [36/50] train_loss: 0.0023, val_loss: 0.0054, val_r2: 0.9946, lr: 0.000115, 70.1s
2023-09-04 19:22:11 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [37/50] train_loss: 0.0022, val_loss: 0.0046, val_r2: 0.9953, lr: 0.000107, 65.7s
2023-09-04 19:23:16 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [38/50] train_loss: 0.0021, val_loss: 0.0051, val_r2: 0.9949, lr: 0.000099, 65.0s
2023-09-04 19:24:25 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [39/50] train_loss: 0.0020, val_loss: 0.0047, val_r2: 0.9952, lr: 0.000091, 68.9s
2023-09-04 19:25:33 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [40/50] train_loss: 0.0021, val_loss: 0.0046, val_r2: 0.9954, lr: 0.000082, 67.2s
2023-09-04 19:26:39 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [41/50] train_loss: 0.0018, val_loss: 0.0050, val_r2: 0.9950, lr: 0.000074, 65.2s
2023-09-04 19:27:44 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [42/50] train_loss: 0.0017, val_loss: 0.0046, val_r2: 0.9953, lr: 0.000066, 65.2s
2023-09-04 19:28:50 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [43/50] train_loss: 0.0016, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000058, 66.1s
2023-09-04 19:29:54 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [44/50] train_loss: 0.0016, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000049, 63.2s
2023-09-04 19:31:00 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [45/50] train_loss: 0.0015, val_loss: 0.0044, val_r2: 0.9956, lr: 0.000041, 66.3s
2023-09-04 19:32:06 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [46/50] train_loss: 0.0016, val_loss: 0.0047, val_r2: 0.9953, lr: 0.000033, 65.5s
2023-09-04 19:33:12 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [47/50] train_loss: 0.0014, val_loss: 0.0047, val_r2: 0.9953, lr: 0.000025, 66.2s
2023-09-04 19:34:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [48/50] train_loss: 0.0013, val_loss: 0.0045, val_r2: 0.9954, lr: 0.000016, 66.8s
2023-09-04 19:35:24 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [49/50] train_loss: 0.0013, val_loss: 0.0042, val_r2: 0.9958, lr: 0.000008, 65.2s
2023-09-04 19:36:28 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [50/50] train_loss: 0.0013, val_loss: 0.0043, val_r2: 0.9957, lr: 0.000000, 64.0s
2023-09-04 19:36:30 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-04 19:36:35 | unimol/models/nnmodel.py | 129 | INFO | Uni-Mol(QSAR) | fold 3, result {'r2': 0.9957527174453936, 'mae': 0.059466276, 'pearsonr': 0.9979272804072913, 'spearmanr': 0.9974385157147599, 'mse': 0.007279518}
2023-09-04 19:36:35 | unimol/models/unimol.py | 114 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /opt/conda/lib/python3.8/site-packages/unimol-0.0.2-py3.8.egg/unimol/weights/mol_pre_all_h_220816.pt
2023-09-04 19:37:41 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [1/50] train_loss: 0.1902, val_loss: 0.1165, val_r2: 0.8797, lr: 0.000267, 65.1s
2023-09-04 19:38:48 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [2/50] train_loss: 0.0543, val_loss: 0.0351, val_r2: 0.9634, lr: 0.000396, 66.5s
2023-09-04 19:39:52 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [3/50] train_loss: 0.0259, val_loss: 0.0187, val_r2: 0.9806, lr: 0.000388, 63.4s
2023-09-04 19:40:57 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [4/50] train_loss: 0.0194, val_loss: 0.0160, val_r2: 0.9834, lr: 0.000379, 63.8s
2023-09-04 19:42:02 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [5/50] train_loss: 0.0176, val_loss: 0.0225, val_r2: 0.9766, lr: 0.000371, 64.2s
2023-09-04 19:43:07 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [6/50] train_loss: 0.0151, val_loss: 0.0120, val_r2: 0.9876, lr: 0.000363, 65.4s
2023-09-04 19:44:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [7/50] train_loss: 0.0133, val_loss: 0.0108, val_r2: 0.9888, lr: 0.000355, 70.9s
2023-09-04 19:45:24 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [8/50] train_loss: 0.0129, val_loss: 0.0108, val_r2: 0.9888, lr: 0.000346, 65.0s
2023-09-04 19:46:31 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [9/50] train_loss: 0.0118, val_loss: 0.0100, val_r2: 0.9896, lr: 0.000338, 66.4s
2023-09-04 19:47:39 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [10/50] train_loss: 0.0100, val_loss: 0.0093, val_r2: 0.9904, lr: 0.000330, 67.6s
2023-09-04 19:48:43 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [11/50] train_loss: 0.0102, val_loss: 0.0106, val_r2: 0.9890, lr: 0.000322, 62.4s
2023-09-04 19:49:46 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [12/50] train_loss: 0.0094, val_loss: 0.0090, val_r2: 0.9907, lr: 0.000313, 62.7s
2023-09-04 19:50:53 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [13/50] train_loss: 0.0081, val_loss: 0.0093, val_r2: 0.9904, lr: 0.000305, 66.8s
2023-09-04 19:51:57 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [14/50] train_loss: 0.0079, val_loss: 0.0080, val_r2: 0.9917, lr: 0.000297, 64.3s
2023-09-04 19:53:04 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [15/50] train_loss: 0.0076, val_loss: 0.0078, val_r2: 0.9919, lr: 0.000289, 65.7s
2023-09-04 19:54:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [16/50] train_loss: 0.0068, val_loss: 0.0080, val_r2: 0.9918, lr: 0.000280, 65.2s
2023-09-04 19:55:15 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [17/50] train_loss: 0.0063, val_loss: 0.0078, val_r2: 0.9919, lr: 0.000272, 65.4s
2023-09-04 19:56:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [18/50] train_loss: 0.0067, val_loss: 0.0099, val_r2: 0.9898, lr: 0.000264, 62.9s
2023-09-04 19:57:21 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [19/50] train_loss: 0.0055, val_loss: 0.0080, val_r2: 0.9917, lr: 0.000256, 62.7s
2023-09-04 19:58:25 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [20/50] train_loss: 0.0055, val_loss: 0.0064, val_r2: 0.9934, lr: 0.000247, 64.2s
2023-09-04 20:02:46 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [24/50] train_loss: 0.0042, val_loss: 0.0056, val_r2: 0.9941, lr: 0.000214, 65.8s
2023-09-04 20:03:52 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [25/50] train_loss: 0.0040, val_loss: 0.0062, val_r2: 0.9936, lr: 0.000206, 65.2s
val:  14%|█▍        | 22/159 [00:00<00:04, 30.91it/s, Epoch=Epoch 26/50, loss=0.0053]              IOPub message rate exceeded.
The Jupyter server will temporarily stop sending output
to the client in order to avoid crashing it.
To change this limit, set the config variable
`--ServerApp.iopub_msg_rate_limit`.

Current values:
ServerApp.iopub_msg_rate_limit=1000.0 (msgs/sec)
ServerApp.rate_limit_window=3.0 (secs)

2023-09-04 20:07:06 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [28/50] train_loss: 0.0035, val_loss: 0.0053, val_r2: 0.9945, lr: 0.000181, 63.6s
2023-09-04 20:08:09 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [29/50] train_loss: 0.0034, val_loss: 0.0057, val_r2: 0.9941, lr: 0.000173, 62.9s
2023-09-04 20:09:18 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [30/50] train_loss: 0.0031, val_loss: 0.0051, val_r2: 0.9947, lr: 0.000165, 68.4s
2023-09-04 20:10:31 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [31/50] train_loss: 0.0030, val_loss: 0.0049, val_r2: 0.9950, lr: 0.000157, 72.6s
2023-09-04 20:11:42 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [32/50] train_loss: 0.0027, val_loss: 0.0057, val_r2: 0.9941, lr: 0.000148, 70.5s
2023-09-04 20:12:50 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [33/50] train_loss: 0.0032, val_loss: 0.0055, val_r2: 0.9943, lr: 0.000140, 67.8s
2023-09-04 20:13:53 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [34/50] train_loss: 0.0024, val_loss: 0.0048, val_r2: 0.9950, lr: 0.000132, 63.0s
2023-09-04 20:14:58 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [35/50] train_loss: 0.0024, val_loss: 0.0045, val_r2: 0.9953, lr: 0.000124, 64.7s
2023-09-04 20:16:03 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [36/50] train_loss: 0.0023, val_loss: 0.0054, val_r2: 0.9944, lr: 0.000115, 64.3s
2023-09-04 20:17:09 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [37/50] train_loss: 0.0021, val_loss: 0.0045, val_r2: 0.9953, lr: 0.000107, 65.6s
2023-09-04 20:18:15 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [38/50] train_loss: 0.0021, val_loss: 0.0047, val_r2: 0.9951, lr: 0.000099, 66.3s
2023-09-04 20:19:19 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [39/50] train_loss: 0.0021, val_loss: 0.0046, val_r2: 0.9952, lr: 0.000091, 63.6s
2023-09-04 20:20:23 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [40/50] train_loss: 0.0019, val_loss: 0.0051, val_r2: 0.9947, lr: 0.000082, 64.6s
2023-09-04 20:21:36 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [41/50] train_loss: 0.0018, val_loss: 0.0044, val_r2: 0.9955, lr: 0.000074, 72.3s
2023-09-04 20:22:41 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [42/50] train_loss: 0.0017, val_loss: 0.0042, val_r2: 0.9956, lr: 0.000066, 64.7s
2023-09-04 20:23:45 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [43/50] train_loss: 0.0016, val_loss: 0.0045, val_r2: 0.9953, lr: 0.000058, 63.4s
2023-09-04 20:24:49 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [44/50] train_loss: 0.0016, val_loss: 0.0043, val_r2: 0.9955, lr: 0.000049, 64.0s
2023-09-04 20:25:55 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [45/50] train_loss: 0.0016, val_loss: 0.0041, val_r2: 0.9957, lr: 0.000041, 65.8s
2023-09-04 20:27:01 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [46/50] train_loss: 0.0015, val_loss: 0.0042, val_r2: 0.9957, lr: 0.000033, 65.3s
2023-09-04 20:28:04 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [47/50] train_loss: 0.0014, val_loss: 0.0046, val_r2: 0.9953, lr: 0.000025, 63.3s
2023-09-04 20:29:07 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [48/50] train_loss: 0.0014, val_loss: 0.0042, val_r2: 0.9957, lr: 0.000016, 62.8s
2023-09-04 20:30:10 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [49/50] train_loss: 0.0013, val_loss: 0.0042, val_r2: 0.9956, lr: 0.000008, 63.1s
2023-09-04 20:31:21 | unimol/tasks/trainer.py | 169 | INFO | Uni-Mol(QSAR) | Epoch [50/50] train_loss: 0.0013, val_loss: 0.0041, val_r2: 0.9957, lr: 0.000000, 71.6s
2023-09-04 20:31:24 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-04 20:31:29 | unimol/models/nnmodel.py | 129 | INFO | Uni-Mol(QSAR) | fold 4, result {'r2': 0.995722167871576, 'mae': 0.058697157, 'pearsonr': 0.997918921152862, 'spearmanr': 0.9975153439068033, 'mse': 0.007057018}
2023-09-04 20:31:29 | unimol/models/nnmodel.py | 144 | INFO | Uni-Mol(QSAR) | Uni-Mol metrics score: 
{'r2': 0.9958098504114381, 'mae': 0.05926784082913989, 'pearsonr': 0.9979403073106544, 'spearmanr': 0.997427024789587, 'mse': 0.0071642844934711545}
2023-09-04 20:31:29 | unimol/models/nnmodel.py | 145 | INFO | Uni-Mol(QSAR) | Uni-Mol & Metric result saved!

代码

文本

Step 3: Hyperparameter Tuning (Optional)

Follow the BBBP tutorial for hyperparameter tuning (link: https://nb.bohrium.dp.tech/detail/1511053944)

The optimal hyperparameters are: Lr=0.0004, Batch size=32.

代码

文本

[ ]

lr_ft = [1e-5, 1e-4, 4e-4, 1e-3] # Learning rates

for i in range(len(lr_ft)): # Loop through each learning rate

clf = MolTrain(task='regression',

data_type='molecule',

epochs=60,

batch_size=16,

early_stopping=5,

metrics='r2',

split='random',

save_path='./exp/learning_rate_' + str(lr_ft[i]),

learning_rate=lr_ft[i]

)

clf.fit("./datasets/RedDB_train.csv") # Train the model

# Load the RedDB_train dataset

df_train = pd.read_csv("/bohr/aorfb-7iz1/v2/RedDB_train.csv")

# Convert the clf.data dict to a DataFrame

clf_data_df = pd.DataFrame({'pred': clf.cv_pred.flatten(),

'SMILES': clf.data['smiles']

})

# Join the datasets on the 'SMILES' column

merged_df = pd.merge(df_train, clf_data_df, on='SMILES', how='inner')

# Save the merged data to a CSV file; the file name will be "learning rate + 'RedDB_train_with_predictions.csv'"

merged_df.to_csv('./datasets/lr_' + str(lr_ft[i]) + '_RedDB_train_with_predictions.csv', index=False)

代码

文本

[ ]

# Get the learning rate corresponding to the lowest R2 value

min_r2_lr = lr_ft[r2_ft.index(min(r2_ft))]

bs_ft = [8, 16, 32, 64] # Batch sizes

for i in range(len(bs_ft)): # Loop through each batch size

clf = MolTrain(task='regression',

data_type='molecule',

epochs=60,

batch_size=bs_ft[i],

early_stopping=5,

metrics='r2',

split='random',

save_path='./batch_size' + str(bs_ft[i]),

learning_rate=min_r2_lr # Use the learning rate corresponding to the lowest R2 value

)

clf.fit('./RedDB_train.csv') # Train the model

# Load the RedDB_train dataset

df_train = pd.read_csv("/bohr/aorfb-7iz1/v2/RedDB_train.csv")

# Convert the clf.data dict to a DataFrame

clf_data_df = pd.DataFrame({'pred': clf.cv_pred.flatten(),

'SMILES': clf.data['smiles']

})

# Join the datasets on the 'SMILES' column

merged_df = pd.merge(df_train, clf_data_df, on='SMILES', how='inner')

# Save the merged data to a CSV file; the file name will be "batch size + 'RedDB_train_with_predictions.csv'"

merged_df.to_csv('./bs_' + str(bs_ft[i]) + '_RedDB_train_with_predictions.csv', index=False)

代码

文本

Step 4: Load the Test Set for Solubility Prediction

The test set has the same data format as the training set. The best training weights have already been saved in /bohr/aorfb-7iz1/v2/best_exp, and you can directly run the following code.

代码

文本

[2]

clf_ll = MolPredict(load_model='/bohr/aorfb-7iz1/v2/best_exp')

test_pred = clf_ll.predict('/bohr/aorfb-7iz1/v2/RedDB_test.csv')

2023-09-07 13:43:56 | unimol/data/conformer.py | 62 | INFO | Uni-Mol(QSAR) | Start generating conformers...
6324it [00:36, 174.39it/s]
2023-09-07 13:44:32 | unimol/data/conformer.py | 66 | INFO | Uni-Mol(QSAR) | Failed to generate conformers for 0.00% of molecules.
2023-09-07 13:44:32 | unimol/data/conformer.py | 68 | INFO | Uni-Mol(QSAR) | Failed to generate 3d conformers for 0.03% of molecules.
2023-09-07 13:44:33 | unimol/models/unimol.py | 114 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /opt/conda/lib/python3.8/site-packages/unimol-0.0.2-py3.8.egg/unimol/weights/mol_pre_all_h_220816.pt
2023-09-07 13:44:34 | unimol/models/nnmodel.py | 154 | INFO | Uni-Mol(QSAR) | start predict NNModel:unimolv1
2023-09-07 13:44:37 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-07 13:44:54 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-07 13:45:08 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-07 13:45:23 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-07 13:45:37 | unimol/tasks/trainer.py | 213 | INFO | Uni-Mol(QSAR) | load model success!
2023-09-07 13:45:51 | unimol/predict.py | 66 | INFO | Uni-Mol(QSAR) | final predict metrics score: 
{'r2': 0.996431347888573, 'mae': 0.05341669959421024, 'pearsonr': 0.9982373718024908, 'spearmanr': 0.99800209666078, 'mse': 0.006212046942459196}

代码

文本

Step 5: Save Training and Test Data as CSV Files (Optional)

You can directly use /bohr/aorfb-7iz1/v2/RedDB_test_with_predictions.csv and /bohr/aorfb-7iz1/v2/RedDB_train_with_predictions.csv.

代码

文本

[3]

import pandas as pd

import numpy as np

# Load the dataset

df = pd.read_csv('/bohr/aorfb-7iz1/v2/RedDB_test.csv')

# Add the test_pred array as a new column to the DataFrame

df['test_pred'] = test_pred

# Save the updated DataFrame as a new CSV file

df.to_csv('./RedDB_test_with_predictions.csv', index=False)

代码

文本

[6]

# Save the training set prediction results along with the original TARGET values as a CSV file

# Load the RedDB_train dataset

df_train = pd.read_csv("/bohr/aorfb-7iz1/v2/RedDB_train.csv")

# Convert the clf.data dictionary to a DataFrame

clf_data_df = pd.DataFrame({'pred': clf.cv_pred.flatten(),

'SMILES': clf.data['smiles']

})

# Merge the datasets on the 'SMILES' column

merged_df = pd.merge(df_train, clf_data_df, on='SMILES', how='inner')

# Save the merged DataFrame as a new CSV file

merged_df.to_csv('./RedDB_train_with_predictions.csv', index=False)

代码

文本

Step 6: Visualization

Visualize the results of our model training by plotting the experimental values against the predicted values, comparing the experimental and predicted values from the test set.

代码

文本

[3]

import pandas as pd

import matplotlib.pyplot as plt

# Read data from CSV files

# train_data = pd.read_csv("./RedDB_train_with_predictions.csv")

# test_data = pd.read_csv("./RedDB_test_with_predictions.csv")

train_data = pd.read_csv("/bohr/aorfb-7iz1/v2/RedDB_train_with_predictions.csv")

test_data = pd.read_csv("/bohr/aorfb-7iz1/v2/RedDB_test_with_predictions.csv")

# Extract the required columns

y_true_train = train_data['TARGET']

y_predict_train = train_data['pred']

y_true_test = test_data['TARGET']

y_predict_test = test_data['test_pred']

# Create a scatter plot

plt.figure(figsize=(8, 8))

# Scatter plot for the train dataset

plt.scatter(y_true_train, y_predict_train, color='orange', label='Train Set')

# Scatter plot for the test dataset

plt.scatter(y_true_test, y_predict_test, color='blue', label='Test Set')

# Set the legend

plt.legend()

plt.xlabel('Actual Target Values of Electroactive Molecules')

plt.ylabel('Predicted Target Values of Electroactive Molecules')

plt.title('Actual vs Predicted Target Values of Electroactive Molecules')

# Set the range for x and y axes

plt.xlim(-7.5, 7.5)

plt.ylim(-7.5, 7.5)

# Add a dashed line for y=x

plt.plot([-7.5, 7.5], [-7.5, 7.5], 'k--')

plt.show()

代码

文本

Step 7. Visualization of Atomic Vector Representations Based on Uni-Mol

Principle: The goal of the following code is to use the high-dimensional vectors representing atomic information in UniMolRepr. By calculating the L2 norm of these vectors, a single numerical value can be obtained, which to some extent integrates all the attribute information of the atom. The L2 norm is a function that maps vectors from a vector space to non-negative real numbers and is used to measure the length or size of a vector. The formula is as follows:

$∣∣ x ∣ ∣_{2} = x 1^{2} + x 2^{2} + ... + x n^{2}$

Next, we further utilize rdMolDraw2D from rdkit to map each atom's weight to a color.

Run the following commands in order in the command line:

sudo apt-get update

sudo apt-get install libxrender-dev 

apt install libxext6

代码

文本

[6]

!sudo apt-get update

!sudo apt-get install libxrender-dev -y

!apt install libxext6

Hit:1 http://archive.ubuntu.com/ubuntu focal InRelease                         
Hit:2 http://security.ubuntu.com/ubuntu focal-security InRelease               
Hit:3 http://archive.ubuntu.com/ubuntu focal-updates InRelease          
Hit:4 http://archive.ubuntu.com/ubuntu focal-backports InRelease        
Hit:5 https://deb.nodesource.com/node_18.x focal InRelease
Reading package lists... Done
Reading package lists... Done
Building dependency tree       
Reading state information... Done
libxrender-dev is already the newest version (1:0.9.10-1).
0 upgraded, 0 newly installed, 0 to remove and 172 not upgraded.
Reading package lists... Done
Building dependency tree       
Reading state information... Done
The following NEW packages will be installed:
  libxext6
0 upgraded, 1 newly installed, 0 to remove and 172 not upgraded.
Need to get 29.1 kB of archives.
After this operation, 113 kB of additional disk space will be used.
Get:1 http://archive.ubuntu.com/ubuntu focal/main amd64 libxext6 amd64 2:1.3.4-0ubuntu1 [29.1 kB]
Fetched 29.1 kB in 6s (4681 B/s)                                               m
debconf: unable to initialize frontend: Dialog
debconf: (No usable dialog-like program is installed, so the dialog based frontend cannot be used. at /usr/share/perl5/Debconf/FrontEnd/Dialog.pm line 76, <> line 1.)
debconf: falling back to frontend: Readline

78Selecting previously unselected package libxext6:amd64.
(Reading database ... 37487 files and directories currently installed.)
Preparing to unpack .../libxext6_2%3a1.3.4-0ubuntu1_amd64.deb ...
7Progress: [  0%] [..........................................................] 87Progress: [ 20%] [###########...............................................] 8Unpacking libxext6:amd64 (2:1.3.4-0ubuntu1) ...
7Progress: [ 40%] [#######################...................................] 8Setting up libxext6:amd64 (2:1.3.4-0ubuntu1) ...
7Progress: [ 60%] [##################################........................] 87Progress: [ 80%] [##############################################............] 8Processing triggers for libc-bin (2.31-0ubuntu9.7) ...
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libcuda.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-nvvm.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvcuvid.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-cfg.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-opencl.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-allocator.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-ml.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-encode.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-opticalflow.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libcudadebugger.so.1 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libcudadebugger.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-ptxjitcompiler.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libvdpau_nvidia.so.525.89.02 is empty, not checked.
/sbin/ldconfig.real: File /lib/x86_64-linux-gnu/libnvidia-compiler.so.525.89.02 is empty, not checked.

78

代码

文本

[7]

import torch as th

from unimol import UniMolRepr

import pandas as pd

from rdkit import Chem

from rdkit.Chem import Draw, rdMolDescriptors

from rdkit.Chem.Draw import SimilarityMaps, rdMolDraw2D

from matplotlib import cm

import matplotlib.pyplot as plt

import numpy as np

import os

import io

import torch.nn as nn

from PIL import Image as PILImage

from matplotlib.colorbar import ColorbarBase

from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas

from matplotlib.figure import Figure

# 2) Define a function to draw atoms, with the data path as input

def draw_atom(data_path):

# Read CSV data and extract the values from the SMILES column

data = pd.read_csv(data_path, header=0)

smiles = data.SMILES.values.tolist()

# Instantiate a UniMolRepr object

clf = UniMolRepr(data_type='molecule')

# Get the molecular representations from the SMILES

reprs = clf.get_repr(smiles)

# Get the atomic representations

atomic_reprs = reprs['atomic_reprs']

# Calculate the norm of each atomic representation as weights

atomic_weights = [th.norm(th.tensor(x), dim=1) for x in reprs["atomic_reprs"]]

# Convert SMILES to molecule objects

mols = [Chem.MolFromSmiles(smi) for smi in smiles]

# Create a Softmax object

soft = nn.Softmax()

# For each molecule, draw a colored molecular image based on atomic weights

for i, mol in enumerate(mols):

# Get the atomic weights and convert them to values in the 0-1 range for mapping to colors

weights = atomic_weights[i]

weights = soft(weights).numpy().tolist()

# Create a mapping from atoms to colors; multiplying by 10000 concentrates all colors within a small part of the plasma color map range, making color differences less discernible;

# By using a larger scaling factor, the weights can be spread over a broader range, making the color differences more noticeable.

atom_colors = {j: cm.plasma(weights[j] * 10000) for j in range(mol.GetNumAtoms())}

# Use the rdkit library's function to draw the molecular image

d = rdMolDraw2D.MolDraw2DCairo(500, 500)

rdMolDraw2D.PrepareAndDrawMolecule(d, mol, highlightAtoms=range(mol.GetNumAtoms()),

highlightAtomColors=atom_colors)

d.FinishDrawing()

# Convert the drawn image to PNG data

png_data = d.GetDrawingText()

stream = io.BytesIO(png_data)

# Use the PIL library to open the image data and save it

image = PILImage.open(stream)

image.save(f'molecule_{i}.png')

# Generate and save a color bar image

fig, ax = plt.subplots(figsize=(0.15, 3.5))

plt.subplots_adjust(left=0.5, right=2)

norm = plt.Normalize(vmin=0, vmax=1)

cmap = cm.plasma

cb = ColorbarBase(ax, cmap=cmap, norm=norm, orientation='vertical')

cb.set_ticks([0, 1])

cb.set_ticklabels(['Low', 'High'])

for label in cb.ax.get_yticklabels():

label.set_family('sans-serif')

label.set_fontsize(15)

plt.savefig(f'colorbar_{i}.png', bbox_inches='tight')

plt.close(fig)

# Open the molecule image and color bar image

mol_image = PILImage.open(f'molecule_{i}.png')

cb_image = PILImage.open(f'colorbar_{i}.png')

# Crop the molecule image to remove top blank space

bbox = mol_image.getbbox()

mol_image_cropped = mol_image.crop(bbox)

# Combine the cropped molecule image with the color bar

combined = PILImage.new('RGBA', (mol_image_cropped.width + cb_image.width, max(mol_image_cropped.height, cb_image.height)))

combined.paste(mol_image_cropped, (0, 0))

combined.paste(cb_image, (mol_image_cropped.width, (mol_image_cropped.height - cb_image.height) // 2))

# Save the combined image

combined.save(f'combined_{i}.png')

# Display the combined image

display(combined)

# Run the draw_atom function in the main function

if __name__ == "__main__":

data_path = "/bohr/aorfb-7iz1/v2/RedDB_small.csv"

draw_atom(data_path)

代码

文本

Step 8: Results and Discussion

From the analysis of the above results, it can be seen that Uni-Mol performs well, with an R² above 0.996. This indicates that Uni-Mol can accurately predict the solubility of organic redox-active molecules from SMILES text data. Further analysis, combined with visualized molecular diagrams, was conducted to examine the impact of functional group types, bonding positions, and quantities on atomic weights. It was found that Uni-Mol assigns relatively lower weight values to hydrophilic functional groups such as —OH, —NH2, —COOH, and —SO3H. Interestingly, even the same type of functional group can have different weight values depending on its bonding position.

代码

文本

reference

[1]Fei Ai, Zengyue Wang, Nien-Chu Lai,, et al. Heteropoly acid negolytes for high-power-density aqueous redox flow batteries at low temperatures, 2022, Nature Energy

[2]Hu B, DeBruler C, Rhodes Z, et al. Long-Cycling Aqueous Organic Redox Flow Battery (AORFB) toward Sustainable and Safe Energy Storage. 2017, Journal of the American Chemical Society.

[3]Sorkun, E., Zhang, Q., Khetan, A. et al. RedDB, a computational database of electroactive molecules for aqueous redox flow batteries. Sci Data 9, 718 (2022).

[4]Sorkun, M. C., Koelman, J. V. A. & Er, S. Pushing the limits of solubility prediction via quality-oriented data selection. iScience 24, 101961 (2021).

[5]Sorkun, M. C., Khetan, A. & Er, S. Aqsoldb, a curated reference set of aqueous solubility and 2d descriptors for a diverse set of compounds. Scientific data 6, 1–8 (2019).

[6]Huuskonen, J. Estimation of aqueous solubility for a diverse set of organic compounds based on molecular topology. Journal of Chemical Information and Computer Sciences 40, 773–777 (2000).

代码

文本

Uni-Mol

点个赞吧