Related Experiment Video
Updated: Feb 19, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
10.0K
Beyond the hype: deep neural networks outperform established methods using a ChEMBL bioactivity benchmark set
Eelke B Lenselink1, Niels Ten Dijke2, Brandon Bongers1
1Division of Medicinal Chemistry, Drug Discovery and Safety, Leiden Academic Centre for Drug Research, Leiden University, P.O. Box 9502, 2300 RA, Leiden, The Netherlands.
Journal of Cheminformatics
|November 1, 2017
Summary
Deep learning models significantly outperform traditional machine learning methods in predicting drug-target interactions. Proteochemometric and multi-task learning approaches further enhance predictive accuracy for drug discovery.
Area of Science:
- Computational chemistry and cheminformatics
- Machine learning in drug discovery
- Bioactivity data analysis
Background:
- Increasing availability of public bioactivity data has spurred chemogenomic research.
- Numerous predictive methods exist, but comparisons are hindered by varied datasets and validation strategies.
- Standardized evaluation is crucial for assessing and advancing computational drug discovery tools.
Purpose of the Study:
- To standardize the evaluation of diverse machine learning algorithms for bioactivity prediction.
- To compare the performance of various methods, including Deep Neural Networks (DNNs), using a unified dataset and metrics.
- To assess the impact of proteochemometric (PCM) and multi-task learning on predictive performance.
Main Methods:
- Utilized a standardized dataset from ChEMBL for evaluating machine learning models.
- Compared Naïve Bayes, Random Forests, Support Vector Machines, Logistic Regression, and DNNs.
- Employed Quantitative Structure Activity Relationship (QSAR) and proteochemometric (PCM) approaches with random split and temporal validation.
Main Results:
- Deep Neural Networks (DNNs) demonstrated superior performance compared to conventional methods.
- The best performing model, 'DNN_PCM', achieved performance significantly above the mean.
- Multi-task and PCM implementations of DNNs improved performance over single-task DNNs; target prediction performed poorly.
Conclusions:
- DNNs represent a valuable advancement over traditional methods for bioactivity prediction.
- Standardized evaluation protocols and datasets are essential for reliable comparison of computational methods.
- The study provides a benchmark dataset and protocols for future machine learning algorithm development in drug discovery.
