相关概念视频
Propagation of Uncertainty from Random Error
Uncertainty: Overview
Propagation of Uncertainty from Systematic Error
Uncertainty: Confidence Intervals
The Uncertainty Principle
Uncertainty in Measurement: Accuracy and Precision
您也可能阅读
相关文章
通过共同作者、期刊和引用图与本文相关的文章。
Enhancing molecular property prediction through data integration and consistency assessment.
Practically Significant Method Comparison Protocols for Machine Learning in Small Molecule Drug Discovery.
UNIQUE: A Framework for Uncertainty Quantification Benchmarking.
DeepCt: Predicting Pharmacokinetic Concentration-Time Curves and Compartmental Models from Chemical Structure Using Deep Learning.
Application of machine learning models for property prediction to targeted protein degraders.
QSAR in the Browser: An Interactive Cheminformatics Web Application.
FoldDoF: Utilizing the Primary Degrees of Freedom of Protein Backbone for Geometric Modeling and Generation.
Derisking Affinity Optimization for Macrocycles and Cyclic Peptides: High-Precision Free Energy Simulations across Five Diverse Targets.
An End-User Audit of Reproducibility, Data Leakage, and Overfitting of the Top-Ranked ADMET Prediction Models in TDC Leaderboards.
PFASGroups: An Open-Source Framework for Automated Identification, Structural Classification, and Prioritization of Per- and Polyfluoroalkyl Substances.
DeepKbhb: Context-Aware Prediction of Human Lysine β-Hydroxybutyrylation Sites.
分子机器学习中的不确定性量化用于数据转移下的属性预测.
Raquel Parrondo-Pizarro1,2, Jessica Lanini1, Raquel Rodríguez-Pérez1
1Novartis Biomedical Research, Novartis Campus, Basel 4002, Switzerland.
机器学习 (ML) 模型预测药物特性,但量化预测不确定性是关键. 将数据和基于模型的不确定性指标与错误模型相结合,可以显著提高分子性质预测的可靠性.
更多相关视频
11:22Using Three-color Single-molecule FRET to Study the Correlation of Protein Interactions
Published on: January 30, 2018
12:05A Simple, Robust, and High Throughput Single Molecule Flow Stretching Assay Implementation for Studying Transport of Molecules Along DNA
Published on: October 1, 2017
科学领域:
- 计算化学是一种计算化学.
- 药物发现 药物发现
- 机器学习应用程序 机器学习应用程序
背景情况:
- 机器学习 (ML) 模型对于预测药物发现中的化合物特性至关重要.
- 准确的预测需要量化ML模型输出的不确定性 (UQ).
- 现有的UQ方法在不同的数据集中缺乏一致的优异性能.
研究的目的:
- 为了对各种不确定性量化 (UQ) 策略进行基准测试,以基于ML的吸收,分布,新陈代谢和分泌 (ADME) 属性的预测.
- 使用UNIQUE框架在数据转移下评估UQ方法性能.
- 确定可靠的UQ方法,以可靠地预测分子性质.
主要方法:
- 使用内部和公共数据集对UQ策略进行全面的比较.
- 应用UNIQUE (不确定的定量化与标记) 框架.
- 在各种数据转移场景下评估UQ性能.
主要成果:
- 基于数据 (例如化学距离) 和基于模型 (例如预测方差) 的UQ指标捕捉了互补的不确定性方面.
- 通过预测ML模型错误的错误模型,将各种UQ指标结合起来,可以获得优异的不确定性估计.
- 错误模型表现出稳定性和高质量的不确定性估计,即使数据转移.
结论:
- 结合各种UQ指标和错误建模,提供了一个有希望的策略,以提高分子性质预测的可靠性.
- 标准化的评估设置和数据转移下的评估对于未来的UQ方法开发至关重要.
- 这项工作为推进化学信息学和药物发现领域的UQ提供了基础.
